Damn. Cool stats. And it looks like my intuition was correct. 
Very nice, borschevsky, but just out of curiosity, how did you figure out which knight or rook was which in the final game position?
And it also looks like your worst bet is knights and center pawns. Nice breakdown, borchevsky!
Thankyou, that’s exactly what I was looking for.
(Are you the same borschevsky who used to post on the Grey Labyrinth?)
I used a tool called Chess Query Language. It allows for specifying properties like this, and will generate a list of the matching games. One nice feature it has that I needed for this is that you can tag a piece and then track that individual piece across several positions in the game. It’s relatively straightforward - for example, to find games where the white queen rook survived:
:forany Rook R
(:position :initial $Rook[a1])
(:position :terminal $Rook[a-h1-8])
The capital R is the designator for a white rook. This code will match any game where there is a white rook on a1 in the initial position, and the same rook is somewhere on the board in the final position.
Yep.
Thank you borschevsky, I will have to take a look at that. I am going to have to improve my chess game now that my 7-year-old has signed up for chess lessons at his school.
I remember hearing about the Chess Query Language several years ago when reading some problems at Tim Krabbe’s site, but it’s still not clear to me where you get the databases to run CQL on.
The database I used came with the Chessbase software. The CQL program works with PGN-formatted chess databases. Pretty much any chess program can create & read this data type. Chessbase uses its own data format, but it is able to read and write PGN. I had to translate their database into PGN to use it with CQL.
If you play any online chess, then you’re likely able to save your games in PGN. Also, sites like chessgames.com allow for downloads of PGN files.
Interesting. You’ll notice the slightly better survival stats for white pawns/pieces compared to their black compatriots. For example, the h2 pawn is more than 2.5% as likely to survive in a checkmate game than the h7 pawn. I think the stats show some interesting things about both opening tendencies and the type of play that is used to create checkmate positions.
Yeah. Also in the checkmate games, the Queen has moved up the list a bit, and the h-pawns have moved down a bit. That makes sense to me. The queen is useful when trying to force a mate. Also, there are a number of attacks that involve pushing the h-pawn against the opponent’s king. These are more likely to end up with the pawn off the board, and a mate for one player or the other.
Also the f-pawns are lower on the checkmate list; the f-pawn is often the focal point of mating attacks.
This is probably related to the fact that white wins significantly more often than black. One would expect a correlation between winning and being able to hold onto your pieces.
While borschevsky’s answer is certainly the best we have, I still think it doesn’t take into account Chronos’s point in post #11, because it ignores draws rather than insisting that all games are played to a conclusion, as the OP later specified. I would guess that most endgames played to a conclusion would eventually boil down to king & pawns vs. king & pawns after exchanges, no? Then one of the pawns gets promoted, mops up most or all of the enemy pawns, and mate soon follows. If that is correct, it would mean that starting off as a pawn is an even better choice than borschevsky’s stats suggest, because they perhaps overstate the survival chances of the rooks and queens,
You need to re-read borschevsky’s post. There are two sets of numbers. The second set is only mates; the first set is all games, including draws and resignations. :smack:
But the point is that the vast majority of games are resigned before mate, in very many cases well before any clear mate is available. Good players will resign as soon as they go a piece down. According to borschevsky, over 95% of games end early in this way or are drawn. The OP appeared to want to know what would happen if all games were played to a conclusion, something that happens quite rearely in real life, so those stats don’t give us the full picture.
(OK, I shouldn’t have said “ignores draws”, rather something like “ignores the effect of draws and resignations on the survival rates”.)
I don’t quite understand - we have 85,000+ games that made it to conclusion to analyze. That’s a decent enough sample space.
They are the small subset of atypical real-life games that end in mate (atypical among serious players, anyway). If all the other games had to be played to conclusion (mate, stalemate, draw by repetition), perhaps you would find that is unusual for anything except king and pawns (possibly promoted) to be left standing.
Consider the Lucenaposition. Here, the players would declare a win for white, and the white rook survives. But played to checkmate, the rooks would be dead after …RxR and KxR.
Mm? Without the rule, then a player might lose even though his king is not in checkmate. That’s a pretty big difference, isn’t it?!
Which would be counted as one of the 85,000 games played to checkmate, no?
Actually, I think I understand the objection: these games would never make it to the 85,000 checkmate positions, as they would have resigned long ago. However, without seeing the database and the end positions of the 85,000 “complete” games, it’s kind of hard to make a determination. Are all or a majority of these 85,000 games unusual mid-game checkmates? If so, I’m sure you can limit the query by number of moves or something like that to get more of an idea of endgame checkmates. I personally don’t think it will make THAT much a difference. Pawns and rooks should still be the most popular endgame pieces, I would think.
Well, I’d keep the rule that check must be declared. From there, the only way that someone would lose without being checkmated is if the player isn’t even trying, since you’d have to know that your king was under attack and still do nothing to save him.
I think the sample is fine. Unless you play to checkmate, the question is pointless.
Hell, if someone retires after being down a piece, you’d never get to checkmate. You can’t answer this question without running the game to its ultimate conclusion, even if the experts do it only in their heads.