setnoescapeon, i agree that playing accurately and playing to win (and succeeding) is far from always the same thing. carlsen is the kind of player that might play "objectively weaker" (or more dangerous) moves, in order to maintain winning chances - i.e. to maintain the possibility for his opponent to go wrong.

however, a modern player without a wc title or any soviet/russian titles could be much "stronger" than tal (hypothetically speaking), but not experience the same amount of success simply because the competition has become much tougher.

hence, relative success (= results) does <not> make a good basis for comparison of "absolute strength" (whatever that means) of tal and a modern day player. and like i wrote in the chessninja post - the skill set needed to be successful changes over time, making the entire notion of "absolute strength" rather questionable in the first place. it also follows that <the talents needed> to be successful will change over time, and accordingly the concept of "most talented player in the history of chess" is equally tainted.

personally i don't see much reason to make these inter-era comparisons in the first place, but if one chooses to focus on relative success, one should at least be very conscious about what one does (and what such a comparison says little about). comparison of ratings of different eras is more or less exactly a comparison of relative success and we don't make it say more about absolute skills by making arbitrary "adjustments" to ratings [like sonas does and confuses people by doing].

the pools are/were (very) different, and the skill set needed to be successful has also changed (more or less, depending on how far apart eras are) and ratings say nothing at all about this, they only measure relative success. it follows that the following two claims are <equally wrong>:

1) because carlsen is rated 2813 in 2010, he is slightly stronger than kasparov was when he was rated 2800 in 1990.

2) due to considerable rating inflation, kasparov's 2800 rating of 1990 proves that he was a clearly stronger player back then than carlsen is at 2813 in 2010.

i think there might have been a slight change in the skills needed to be successful in the last 20 years (due to the changes in how people prepare for opponents, for instance, but also due to the wide-spread use of faster time controls, no more adjourned games which result in longer playing sessions amongst other things, and so on). much more important, imho, is that the breadth at the top has become much bigger (which again influences preparation, etc.)

anyway, there's simply no good way to know how carlsen 2010 would've done against kasparov's opponents from 1990, or how kasparov 1990 would've done against carlsen's opponents from 2010. that's how the respective ratings have come into being. note that in this perspective it's <much less relevant> how a very hypothetical match between kasparov 1990 and carlsen 2010 would've ended, imo. head-to-head is both a different way of comparing "strength" and it also requires a different skill set than normal tournament play. the 1990 kasparov would've had much more experience in match play than the 2010 carlsen, for instance. while many people consider head-to-head to be the ultimate way of deciding who the better player is (and to decide the world championship), i think it's only one of several methods of comparing the strength of two players. a match can always be won with a +1 score, for instance - quite fundamentally different from the typical tournament.

I believe we do agree about it's definition.

Rating inflation means an effect which causes a rise in top level FIDE-ratings, even when there is no rise in actual playing strength.

In other words, it's a question of whether there are any inherent properties of the rating system which causes this effect.

I don't know if this effect exists. But, depending on how the model is put together, i do believe it might exist.

In high jump, the advent of a new technique will raise the scores of all contestants, as you say.

In chess, this is not so, because the FIDE rating system is a contained system. That is, points (in theory) do not enter or leave the system. When one player gains a point, another player loses a point.

So, if all have access to new computer based methods, it does not cause the ratings of all those players to rise, like it would in your examples of ski jumping or high jump.

kramer, let me say this one more time: the system/model does <not> have as one of its properties to maintain any relationship between rating and skills. ratings portray results, not skills.

hence, the assumption that the model somehow should keep a non-existing relationship intact is what i consider the problem here, not how the model/system works.

this was exactly what the statement quoted above tried to say something about:

<"[I]f it would sink in that there is NO DIRECT MAPPING between absolute "chess skills"/"quality of chess"/"chess level" and the chess ratings of FIDE (or any other widely used rating system), there would be a chance that rating discussions could get anywhere.">

i went on to explain why i think there can't possibly exist any such (trustworthy) mapping.

<Rating inflation means an effect which causes a rise in top level FIDE-ratings, even when there is no rise in actual playing strength.>

here you use a term that i'm quite sure we disagree about the meaning of: <"a rise in top level fide-ratings">

what does that mean, to you? how do we check for a rise in top level fide-ratings? please try to relate it to the number of strong players in total in the pool, for example by answering these two question:

<how do we discriminate between such "rating inflation" and a potential increase in the number of "very, very strong" players? how would the latter theoretically materialize?>

note that <my> definition of <nominal rating inflation> is something like this: if the average rating in a closed ["randomly" selected] group of [supposedly pre-dominantly stable] players go up over time, then there is rating inflation.

one more time, without the clarifications: <if the average rating in a closed group of players go up over time, then there is nominal rating inflation.>

of course, the group should be of an "appropriate size" in order to be statistically significant. also, the selection of the group can't be such that the normal expectancy would indeed be a notable rating increase (the top 20 finishers of a boys under 14 world championship are expected to improve and raise there ratings, i guess) - or inversely, such that the normal expectancy would be loss of rating points (as if our sample/closed group would consist of all players aged 70 or more).

It's not necessarily non-existing.

If you look at the whole group of players, it's conceivable that the intelligence / skill of the whole group is about the same now, as it was for instance at the time of Bobby Fisher. And, furthermore, that the distribution of skills follow the same normal distribution as it did then.

If that is the case, then the FIDE-rating, according to it's goals, would indeed say something about skills.

And, if the opposite is true, that todays player generally are more skilled, because of the access to computer training, it means that a 2800-player today is more skilled than a 2800-player was 20 years ago.

The question raised is if there are any other effects, caused by inherent "flaws" in the system, which affects the relationship between skills and ratings.

You know the basis for the calculations better than I do, so I suspect you should be able to provide an answer... :)

I agree, this is a good definition.

Still, it's possible to conceive that there are effects, which affect, not the "average" rating, but mostly the ratings of the very best players.

One such effect could be (as I've speculated earlier) that the number of rated amateur players grow faster than the number of high level players, for some reason.

In other words, that the sport grows in size as a hobby, but stays the same size as a professional sport (for some reason).

If this is true (which I don't know if it is), I imagine that it could have the effect of inflating the ratings of the high level players.

that's quite clearly so, which i've also stated several times: the increase in the pool is largely due to the addition of people like me - or even weaker players with little potential for improvement. but guys like me don't necessarily become much weaker over time either.

[my own fide rating is up about 100 points since i first got it, while the average of all fide-rated players in the pool is down about 100 points, largely due to the lowered lower bound on fide ratings.]

<In other words, that the sport grows in size as a hobby, but stays the same size as a professional sport (for some reason).>

isn't it quite obvious that it has <not> stayed the same as a professional sport? there were less than 600 players in the first fide rating list, for instance. do you think that all of those 600 players would've been 2500+ rated gms today?

<Still, it's possible to conceive that there are effects, which affect, not the "average" rating, but mostly the ratings of the very best players.>

i agree that it's possible to conceive that there are such effects, but i've never seen any effort of explaining such an effect and how it can be tested (and potentially falsified). do you have such a suggestion? the the relative increase of rated amateurs is no such "effect", unless one can demonstrate or postulate the mechanism which makes the top players' ratings go up due to a relatively higher number of rated amateurs.

i'm still very interested in your view regarding these questions, btw:

<how do we discriminate between [...] "rating inflation" and a potential increase in the number of "very, very strong" players? how would the latter theoretically materialize?>

again, why would it be a flaw in the system if there is no fixed (consistent over time) relationship between skills and ratings? no matter what fide says or wants, rating systems exclusively based on results can't make such promises, can they?

<You know the basis for the calculations better than I do, so I suspect you should be able to provide an answer...>

So... if that is your starting point, you miss the point of the discussion.>

isn't it quite obvious that it has <not> stayed the same as a professional sport?>

Sure. I knew that one was coming... :)

I exaggerated the point, in order to make it more clear. The point is the same though, even with some growth in the number of professional players, as long as this growth is smaller than the growth in the number of amateur players.

<i agree that it's possible to conceive that there are such effects, but i've never seen any effort of explaining such an effect and how it can be tested (and potentially falsified). do you have such a suggestion?>

I guess it's hard to prove or to test.

But, maybe it can be looked upon in the following way:

Situation a: Magnus has one 2200 player to play against. He beats him once, sending him down to 2198. He beats him again, sending him down to 2196. Etc. For each win Magnus gains fewer points, because the rating of his opponent is declining.

Situation b: Magnus has two 2200 players to play against. He procedes in the same way, beating both of them several times. In this situation, Magnus will gain more points, because he earns them from two players instead of from one.

It's a crude example, I know.

<how do we discriminate between [...] "rating inflation" and a potential increase in the number of "very, very strong" players? how would the latter theoretically materialize?>

how do you think the above would materialize in the rating system?

As for a rise in the number of very very strong players, that also begs an explanation.

I see two possible explanations:

a) Some of the top players today represent early adopters of computer based chess tools, so they race ahead, opening up a gap to the rest of the pack.

b) The introduction of computer based tools increases the overall complexity of the game, which agains creates a larger distance between those who master it all, and those who don't. In other words that the introduction of computer tools in itself creates a "stretch" in the field, so to speak.

kramer, let me say this one more time: the system/model does <not> have as one of its properties to maintain any relationship between rating and skills. ratings portray results, not skills.>

There will be a rating committee, and I presume they all monitor inflation/deflation in the system. Also, the rating committee can decide on countermeasures to combat inflation/deflation.

<"[I]f it would sink in that there is NO DIRECT MAPPING between absolute "chess skills"/"quality of chess"/"chess level" and the chess ratings of FIDE (or any other widely used rating system), there would be a chance that rating discussions could get anywhere.">

Isn't the USCF rating system widely used?