Which endgames should you practice? A statistical analysis

@ProfDrHack

"Shouldn't the (detailed) percentages add up to 100%?"

No, because many games include more than one type of endgame.

"contradict each other"

No, because one is a share of games, and the other is a share of positions.

40 half moves is OK, I thought it were 40 moves i.e. 80 half moves.
Rook endings remain most important, but endgames become more prevalent with the 9 men definitions as opposed to your restricted set.
I still think that the 5 pts material difference is too restrictive and eliminates many technical endgames like Q vs R, Q vs B, Q vs N, Q vs pawns with or without some pawns on either side. There are some important known draw positions and some that are hard to win for the queen.

@tpr Oh sorry, I mistyped: it actually is 80 half-moves. I know it's limiting, but I've already wasted too much time on this to run everything again :)

I like this statistic. And i can confirm that reading through Smyslov/Loewenfish has made me play rook endgames much better.

However, i may add that it is not just how often the endgame will happen that makes it important.

For example if one studies pawn versus knight and the typical motifs - he will not have exactly this endgame often, but the motifs – for example, pawn attacking knight on eight rank promotes, or knight stops pawn by giving check at the same time – will also occur in middlegames.

Or if one is aware of the fact that the queen is the only piece which can bring through the pawn alone without help by king - he will also not have that specific endgame often, but it for example helps him in middlegames with a passer, he will be able to figure out if he should keep queens on board or not.

So endgame study is good in general for the same reason why doing tactics is good, because one learns typical motifs, which are useful in any stage of the game.

Addon: Also the fact that the endgame does not occur often does not mean that it was not a possibility in the game. The player may have decided to avoid it based on his knowledge about this endgame!

Ok I understand the > 40 move = 80 ply limit. That is Ok.
Is it possible to lift the < 5 material difference limit so as to include endgames of Q vs piece or Q vs pawns?
What becomes then the occurrence rate of endgames defined as 9 men or less? It presumably becomes more than the 14% above.

@theoriefritze

Your points all sound reasonable in principle. Basically, the argument for studying rare endgames boils down to an argument about "transferable skills".

My intuition is that skill transfer is easier when the two situations are similar. So if I want to improve my middle game, it makes more sense to study middle game-looking positions. But that's just intuition.

I'm sure a bunch of psychologists have already studied all that...

Your addon is also interesting. I would just note that the main motivation for this was that I wanted to know what the typical amateur's games looked like. You're rated better than 99% of Lichess players, so I can see you making that kind of judgement. For reference, my peak rating is 1900 (92nd percentile), and I rarely (ever?) make judgements about a position using that kind of reasoning. I know that's a flaw in my game, but still, I'm not sure the selection problem is too huge in that sample.

@dudeski_robinson Actually the difference between opening, middlegame, endgame, (and also tactics and strategy) is quite theoretical in my opinion. It is all about knowing motifs and awareness in chess. It does not really matter if one studies the motifs with middlegame positions or with endgame positions, i would guess both leads to the same result.

@theoriefritze I'll defer to the opinion of the stronger player here :)

@tpr When I raise the limit to 9 pieces per player and <8 in material difference:

R: 77% of games
Q: 44%
B: 49%
N: 49%
B (no R): 18%
N (no R): 21%

+1 to everything tpr said, especially >3 pawns and QvR.

I would say <9 men or <13 material each makes the position an endgame.

Not that it would change the fact that most endgames are rook endgames, as most players can check just by looking at their game history.

@tpr I find it hard to make a case for RB+6p v RN+5p being an endgame. I agree it may well be so, but for automatic detection purposes, there are still many pieces, and much material. Middlegame principles may well be more applicable than endgame techniques. (Also, then what about RR+3p vs RR+2p, is that also an endgame ?)