Chess Tournament Games and Elo Ratings (2014) | Hacker News Reader

Chess Tournament Games and Elo Ratings (2014)

randalolson.com

45 pointsdangerman9 comments

Very nice article, and generally meshes well with my experience as an expert player(probably master if I can put in another good run of real training).

If anyone wants a complete deep dive into the quirks of chess ratings, and attempts at modelling the relative strengths of players separated by centuries, this page is wonderful if you can handle a little bit of math and spartan text formatting: http://www.chessmetrics.com/cm/CM2/HistoricalRatings.asp

I also like that the article mentioned my favourite historical match! Tal-Botvinnik 1960 has everything you want in a WC match. The battle between two diametrically opposed chess archetypes and personalities. The positional opening theoretician Botvinnik, academic, patriarch of Soviet chess. Against the barbaric attacking player Tal, a sickly young school teacher from Riga with 3 fingers on one hand and a cigarette always in the other(during the games, at the board, naturally).

Tal wrote a book(titled simply Tal-botvinnik 1960) on the match with his analyses of all the games, which I recommend to anyone interested in great historical chess games. Tal was a really talented writer as well. It's available as an e-book.

Tal is a severely underrated player too. People including himself always said his play was dubious. But he also had the longest unbeaten streak of rated chess in history until Ding Liren beat it a few years ago. But let's just say Ding's run had a lot more draws than Tal's run. Tal was so hard to beat because while he might have dubiously sacrificed half his pieces, his opponents king would be bare and he would always be able to find a perpetual check and make a draw. That is if he didn't checkmate you instead.

Tal is my favorite player and what got me playing chess again. Watching his games is very entertaining with his relentless attacks and sacrifices. He also had a very impressive calculating ability and a lot if not most of his sacrifices were meticulously calculated and proven sound by modern engines.

Another fun and more contemporary player to watch is Vasyl Ivanchuk who had immense talent and could have become world champion if it was not for his nerves.

Yeah, Tal's sacrifices being unsound is more or less a meme and fails take into account the historical context the games were played in. Tal did miss some things in his calculation, but they were usually things the opponent would never find. Because at that time, essentially no one could outcalculate or even match him in complex positions except maybe Viktor Korchnoi(who somehow had a clear positive score against him) and Fischer in his prime(though since he stopped playing in his prime he never managed to get a + score). So if Tal missed a refutation to one of his sacrifices, almost certainly the opponent missed it too. It was usually some bizarre computer line. Sometimes even grandmasters analysing the games at the time didn't find them, so clearly finding them at the board was pretty much impossible.

I never really studied the Karpov-Tal complex of games but somehow all their 23 games but one were a draw, with one win for Karpov. I think this is more a case of Karpov's ability to avoid tactical complications and outplay his opponents positionally than matching his calculation. But interestingly, Tal positional play mostly survived the Karpov test, which is impressive. Tal was also much older than Karpov. His peak ELO was 1705 in 1980 which is nuts considering he was 44 by then and his health was deteriorating, yet he was still behind Karpov in his prime by only 20 ELO points. Probably all those draws with Karpov helped a lot...

In the historical chess ratings linked in GP, Tal's peak rating is 2799 in the 60s. Kasparov peaks close to 2900 in this system though and so does Fischer. Unfortunately, it seems the ratings stopped being updated ariund the mid 00s, so current top players don't have meaningful ratings :( I would love to know if Carlsen et al would come out close to the same as Kasparov and Fischer, or whether they'd still be ahead by a significant amount. Hell, maybe they'd be behind, since they don't necessarily dominate in the same way.

> Elo ratings tend to accurately predict the outcome of a game. For those who are newer to chess, just know this: If you're pitted against someone with a much higher Elo rating, you can expect a quick and decisive defeat.

Isn't that by definition what ELO is? From Wikipedia:

  The difference in the ratings between two players serves as a predictor of the outcome of a match. Two players with equal ratings who play against each other are expected to score an equal number of wins. A player whose rating is 100 points greater than their opponent's is expected to score 64%; if the difference is 200 points, then the expected score for the stronger player is 76%.

https://en.wikipedia.org/wiki/Elo_rating_system

> I'd imagine the only reason this trend levels out at ~90% is because this data set contains games where a talented new player hasn't quite reached their proper Elo rating yet.

Well, first of all it can't go above 100% so it obviously has to level out somewhere, and second, this is _by definition_ what the curve should look like: A 400 point ELO difference is a 10:1 expected score difference.

> Newcomers, take note: Try to play as White as much as possible [...]

...that's kinda like saying "play against lower-rated players to win more". Sure, if number of games you win is the only thing you care about (and not such things as "getting better at the game" or "playing a fair match"), that's a great idea.

Overall the article has interesting plots but very under-informed commentary imo.

Serves as “a” predictor is key here. You can have very good predictors and predictors with very lackluster performance. What deviation in rating gives you say only a one percent chance in winning is an indicator for the quality of the predictor. Think of an hypothetical ELO-rating for chess >> tennis >> soccer. If that’s too trivial, perhaps archery is a nice example. I think ELO is applicable to archery. Which rating is a stronger predictor? Where are upsets more likely? Chess or archery.

Casually eyeballing the results, one thing that stands out to me is that the empirical winning chances for stronger (higher Elo) players are actually better than the "normative" winning chances the Elo algorithm was designed to converge on. I wonder if that pattern holds up under scrutiny? It might be an interesting anomaly to explore, at any rate.

There are some practical quirks to the realities of chess tournaments, and who wants to play who, which I bet lead to a lot of deviation.

In practice, a super GM in their late 30s would hate to be playing up and coming teenagers that just made GM since they are underrated (Those kids gain 200+ points of ELO in a year or two) and the older GM is probably getting worse anyway. So, in practice, there is a difficult transition to get a high enough ELO vs regular GMs and IMs to have good chances to aim for a 2700, where the real gains are at.

Add to that the fact that there are few classical tournaments, while the youth is training in blitz online, and you can find pretty deceptive ELOs out there.

The data are interesting. But almost all of his conclusions and commentaries are pretty dubious.

For instance:

> Therefore, it's in your best interest to play against others around or above your current Elo rating.

No, that doesn't follow. There are certainly valid strategies in constantly beating people rated much worse.

Not if you want to become a stronger player.

Reply on news.ycombinator.com