Here is an interesting visualisation of draw stats looking at both rating level and rating gap between opponents…
web.archive.org/web/20160806071058/http://chess-db.com/public/research/draw_rate.htmlShows a substantially higher rate of drawing at higher levels, and that the draw rate increases steadily all the way up from 1400 to 2600.
(The original site seems not online atm, so this is an archived copy.)
Credit where it’s due… This data was shown to me by my teammate
@chutzpah a couple of years ago when we were building some simulations of winning probability in team matches.
The restricted understanding of op question is about the individual player improving though.
So my database concern, I should try to tie it back:
Well, YES, but, do not use the masters' database as argument to that. it is not yet reliable. if that warning is not off-topic. then the discussion about database argument should be just an off-shoot.
#11. We would need to know the type of data there. A database or population argument needs full disclosure on the database construct. I only trust lichess user-base database for their draws stats (to the extent that tournament results containing matches are negligible proportion, dunno anything about lichess tournaments).
Perhaps that exercise should be remade with lichess. Then the question would be not do draw levels increase, but how much more? What lichess database would show (in quantitative terms), with more confidence because of full open source (almost open data) principles, and access to provenance of the game data (context).
But yes, with direct understanding of the question. The *qualitative* answer is yes. Just not what the master database shows as *quantity*, although it surely makes people ask that question...
The above (#11) would need more information about its data source, to make it more than a qualitative argument. which may be considered off-topic. I wish though for a cleaner master database in the future, no need to discard anything just add one more category of filters, like for lichess. but replacing time control with game length categories. so that one can choose which draws to filter out of the statistics. If masters keep struggling longer, I have more belief in the value of the draw outcome. And I think given the unknown and tournament hypothesis, this point should be part of this thread. done. unless new information or arguments.
I will try to find where the data comes from with the link above.
Draw Rate in Chess Tournaments, it says.... Thanks for the find .
Within tournament lower levels less draw proportion. Maybe also less tournament experience. Is there a way to filter that data with match size, and position of game in math. and ultimately, the Tier story, if that may also play a role (which adversary coming in the future, i don,t know). I will look more into details. time scale.. but if you know about the data context, the more the better. for that point.
The tournament experience, and reduced tier-ing can make the meta-(mono-game) competition clearer.
so tournament information context might still be a factor. as not many low levels will go in the clearer context.
But I am not arguing the trend.. just the database quantitative content, and the reliability of its draw proportions, but most importantly for the learner in me, the association between position and a draw of convenience not being labelled as such.
@jagdip do you know the data source?
I followed the only 2 links there and unfortunately the links are for the data analysis methods, as far as i could detect,
and a link to engine competitions* draw rates.
* perfect chess even if only subset of it, is likely to tend to near draw, how much is white initiative at t=0 going to carry over there?.
Should I explore in that web archive? If you have some pointers to source data that would be nice complement to your post. Ability to answer myself some of above. But I think the are modern ways to test. But not much high level curiosity, and only buffoons like me would insist.
I don’t know more than what is on the page.
8 million games though is in the same ballpark as the Chessbase database, so it would have to be something like nearly every published competitive game ever to have that many games.
en.chessbase.com/post/new-mega-database-2021I would guess they use the word “tournament” loosely to mean competitive published games as opposed to for example casual games or online games.
Btw “matches” - meaning a series of games between the same two players - are going to be a very tiny proportion of competitive chess games.
A bit late, i missed that last post back then. now reviewing stuff I wanted to keep in mind.. And voilà. An order of magnitude clue about matches. (I should find out why some tournaments use those or not, what are the associated context, big or small, high level entry point of not , if that is a thing). But your statement is noted. Thanks. I should find out my past funky ones (to-me), to make sure my match hypothesis applies there (it depends if all of the book ids point to unrelated time scales or an equivalently large group of different pairs. Memo to that.
