The valley of death

An interpretation of this data:

I was a 2153-rated expert who abruptly lost his tactical acumen but who, via herculean effort involving  a decade of intense study of classical openings, positional concepts and endgames, completely remade himself into … a 2153-rated expert.

Um, yay?


19 thoughts on “The valley of death

  1. As I understand it they’re bleeding points into the system. And given that I know a bunch of veterans who are also on or near lifetime highs, it does seem likely that we’re “wind-aided”, as they would say in track.

    But it feels like an accomplishment nonetheless.

  2. The bonus factor is lower than it’s ever been, at 12. But, supposedly, Mark Glickman has inflation under control. Only he knows the current stats.

    MCC players might be a tad over-rated compared to the rest of the country. Having Arthur Bisguier sitting on his floor, at 2200, losing to 1800s doesn’t help MCC rating inflation. If Bisguier donates, say, 50 points per month — 600 points per year — and divide that between the 30 or so regulars in the MCC Open section, that could translate to an average of 20 extra points per player. It helps to beat Bisguier, but second-order effects, based on people who play players who beat Bisguier, are also in the air.

    I’m not saying that Bisguier is the reason, but when a 2200-player is sometimes playing like an 1800, it help local-area inflation. Even 2500-players who beat him are losing less often than they should.

    Just a thought….

    Rating inflation is nice on the ego, but it would be better, I think, if USCF ratings were more in line with FIDE ratings.


  3. re: inflation & FIDE, I agree.

    I was told the bonus factor was raised, but I was evidently misinformed. (*cough*)

    True about floored players, although I wonder if the constant flow of coached kids is a mitigating (not to say completely balancing) factor. i.e. There are a lot of players who are consistently improving and whose ratings thus lag a bit behind their skill level. (Greg K has previously argued this lag isn’t as bad as in the old days, when ratings were updated very slowly.)

  4. USCF ratings may be inflated as compared to FIDE ratings (though consider the interesting counter-example below which I posted in another context at LEP). That will be the case at least as long as the odious concept of a rating “floor” continues to exist (here’s an idea: just keep the floor for honorary and anti-sandbagging purposes but track and use the “actual” rating for doing the rating calculations).

    That said, I have two thoughts on Howard’s and Derek’s comments. First, there are floored players all over the US that are skewing the ratings calculations, so it’s not just an MCC problem (and given its prevalence, I think that it largely comes out in the wash).

    Second, the reason that there are so many players at or near their historical floor cannot just be explained by the advancing age of those players (although in some cases that is of course the dominant or even the sole explanation). I think a big reason is that there are so many online resources today that it is much easier today for weaker players to compete with stronger ones. So, in other words, an 1800 player today is a much tougher out for an expert or master than even 10 years ago when online resources were not as well-developed as they are now. I don’t think Curdo, for example, is spending hours on ICC, working through tactical programs, or fine-tuning his Schliemann or Dutch through Fritz and online databases. As a result, an A-player who is doing those things will have a decent chance of beating or drawing him. I think, therefore, that floored players are not just getting older; I suspect that they are also not adequately availing themselves of these online resources, which places them at a disadvantage against the many who are.

    See where I’m going with this? I think that the typical non-floored 1800 player today is significantly better than a non-floored 1800 player even 10 years ago. And, similarly, I think that a 2153 rated player today is significantly better than a 2153 from 10 years ago. So I think that Derek’s return to 2153 is an extraordinary accomplishment, not only because he returned there from a rating in the 1950’s but also because he has almost certainly become a much stronger player than he was the last time he was 2153.

    6) Are USCF Experts Stronger than Russian IMs?
    It use to be common knowledge that Russian Experts were stronger than American players with USCF ratings of 2400, but has the tide turned? Recently Expert Andrei Blokhin of Maryland (currently rated 2138 USCF/ 2395 FIDE) received the title of International Master in recognition of IM norm performances achieved in round robin events in Moscow in late 2001 and early 2002. Mr. Blokhin, who has played in several Under 2200 sections in World and Chicago Opens without ever winning top prizes is not a sandbagger. His USCF rating, based on plenty of activity, has floated between 2081 and 2167 for the period 1993-2002. Shortly after making his norms he scored 3.5 from 6 against USCF 2100s in Chicago. Does this mean that things have changed and USCF Experts would be 2400 IMs in Russia?

  5. Hi. I have to chime in again, even though this thread is getting old. With all due respect, Greg, today’s 1800 off his floor, relative to, say, a Curdo at 2250 off his floor, has the same 200 point rating difference today as he did 10 or 20 years ago. All players today may play stronger than players of the same rating 20 years ago, but relative differences in strength always — and this is a mathematical conclusion — remain the same.

    What happens sometimes is that a small subset of players inflate or deflate their ratings, if they do not mix well with the rest of the rating pool. If everyone at the MCC somehow got better, all at the same time, and everyone at the MCC only played each other, then everyone’s rating at the club would remain the same. But these players would be underrated compared to the rest of the country.

    I could go on at length about ratings, but I need to get back to work {sigh}…

    One more thing: I asked Glickman about the FIDE-USCF relationship in ratings, back in my October 2006 CL interview. What he said, and it’s in the article, is that USCF players with USCF ratings between about 2000 to 2400 (I forget the exact range) have slightly higher FIDE ratings, and USCF players rated 2400+ have slightly lower FIDE ratings. Notice, I did not say that USCF ratings < FIDE ratings across the board. I said only that they are not aligned. Honestly, I don't know why the weird discrepancy exists, but it does….at least in 2006. The most likely cause is that ratings are not absolute values of strength; they are estimates at best; local fluctuations exist; there is little "cross-pollination" between international players and Americans, etc.


  6. I think historically there is a myth that USCF ratings < FIDE ratings across the board. But Glickman did the math, found the true relationship (circa 2006) and your personal evidence supports the data.

  7. Ooops, I meant to write “USCF ratings > FIDE ratings across the board,” in both posts.

  8. Howard – very interesting, and as you say that is a weird discrepancy above and below 2400.

    This is not the reason I’m reading Statistics for Dummies, but an extra incentive can’t hurt. :) (So far the first chapter is REALLY for dummies.)

  9. Howard — Based on your use of the phrase “with all due respect”, I gather that you disagree strongly with something I said. But I’m not sure I understand what that something is. Perhaps we will show up at the same tournament one of these days and can discuss it in person.

  10. Hi, Greg

    Yes, I did disagree with something, but I did not make myself clear. What I could not understand was your conclusion of how players not on their floor today could be stronger than players of the same rating not on their floor 20 years ago, if so many other players ARE on their floors. The fact that many players may be on their floors has the tendency to inflate all the other non-floored ratings. I implied this in my other post, but I was not completely clear.

    Yes, it is technically possible that all players today are stronger than their same-rating counterparts 20-30 years ago (based on easy access to better training techniques), but I doubt it. I think that it is more likely that players are weaker than their same-rating counterparts, because of inflation and so many people being on their floors. I respect our disagreement. :) I just think that inflation due to floors outweighs deflation due to better training techniques.

    I’m not a statistician, but some people believe that the presence of more top-100 players peaking over 2700 now than 25 years ago is a sign of better training techniques. This may be a real phenomenon. Everyone may be getting better, and it’s showing at the top, with the existence of actual higher top ratings. But that doesn’t mean that a 2200 FIDE player today is better than a 2200 FIDE player from 1980, does it? To me, it just means that the best players of today are better than the best players of 1980. But match a 2200 from 1980 vs. a 2200 from today, and they would score even in a long match.

    I’ve been playing competitive chess for 25 years. My subjective opinion is that USCF ratings have been relatively reliable over that time. Glickman has the official data. I can’t speak for FIDE ratings, because I don’t have one (yet)!

    The one thing that has improved considerably in the past 25 years is the timeliness of the rating calculations. Ratings now get computed within a day of a tournament ending. This, along with a higher K-factor for lower ratings, has totally eliminated underrated kids. Certainly an improvement over yesteryear.


  11. Howard,

    Yes, sounds like we will have to agree to disagree on that one. There is no doubt in my mind that a player rated, say, 2207 today is significantly stronger than a player with the same rating ten or more years ago.

    To understand why I am so convinced of this, one needs only to review my tournament games from 1987-89. I don’t think that there is any reasonable doubt that the player playing those games is approximately 2000 strength. But my rating at the time was in the mid to high 2100s. And I can tell you based on my own subjective views that the Greg K of 2010 (rated 2207, coincidentally enough) would absolutely annihilate the 2177- rated (just to pick a random rating from my old scorebook that I just opened) Greg K of 1988 in a way that would suggest a rating difference between the two players somewhere in the range of 150 to 200 points.

    Obviously this is all subjective and anecdotal but I have an endless supply of anecdotes about other players that only reinforces my conclusion. In the interest of protecting the innocent (and avoiding further clutter from me on Derek’s blog), I will save those anecdotes for another time.


  12. Greg,

    I actually enjoy this conversation very much, and would not mind taking it offline. Shoot me an email if you want (w_rox at yahoo).

    For all practical purposes, your rating of 2010 is the same as it was in 1987. 2207 is only 30 points higher than 2177, not quite statistically significant. You may look at your games and say, “Gee, what an idiot I was back then! How could I make such stupid moves?” But this does not consider that the 2010 Greg has certain weaknesses that the 1987 Greg did not. You’re 23 years older! Maybe you’re much better at certain skills but worse in others.

    I believe that Mark Glickman has tracked historical data for as long as he has been on the USCF Rating Committee. He tracks the average rating for players between ages 30-50 (or some similar age range), and has this data. He found some age range that he believes should maintain a stable mean rating. (This, of course, is not necessarily a valid assumption, but it’s the only work we have.) He tries to keep this mean rating constant over the years, to maintain the credibility of the system.

    Now, the USCF has rating data for every game played, going back to 1990. So it’s conceivable that someone can write a program to analyze this data and come to some conclusions independent from Glickman’s.


  13. Comments are not clutter!

    I just don’t understand why we aren’t having a similarly substantive discussion on the Tube Museum post.

    — There’s no question that Circa 1987 Derek would have CRUSHED Circa 2010 Derek in any tactical position. In fact this kind of exercise

    provides a handy quasi-empirical basis for comparison. I know that back in the day, my calculations were deeper (by maybe 3 to 6 half-moves), wider, quicker, AND less prone to oversight.

    OTOH in 1987 I played some incredible positional howlers that I recall quite vividly – frequently closing the only line that offered me counterplay, making gigantic holes in my pawn structure or near my king, lashing out to lose drawish endgames, etc.

    So it kinda balances out for me, such that I can’t categorically say today’s 2153 is better or worse overall.

Comments are closed.