lichess.org
Donate

Is Carlsen better after losing a game?

Has colour been taken into account? If he is generally alternating colour he plays, and if losses are more likely as Black / wins are more likely as White, this alone could perhaps explain the trend in classical format.

Has colour been taken into account? If he is generally alternating colour he plays, and if losses are more likely as Black / wins are more likely as White, this alone could perhaps explain the trend in classical format.

@DaBassie said in #9:

I was just wondering about your statistical method here. Isn't looking at 'the game after a lose' introducing some statistical bias? Like, Carlsens loses are already pretty rare in classical, and by taking the game after a lose, you're by definition excluding one lose out of your sequence of results (Wins/Draws/Loses). So what remains of this sequence is then guaranteed to be of higher performance then the sequence you started with.

I'm not sure how OP calculated it, (whether its just one game or just rest of tournament- i'm assuming the first one), but that would also mean it excludes prior results as well. for classical it would be biased towards prep work, and carlsen didn't really seem to care as much previously for blitz/rapid iirc, since his main time was going into prepwork for the classical world championships. he just happened to be a title holder of the other two.

@DaBassie said in #9: > I was just wondering about your statistical method here. Isn't looking at 'the game after a lose' introducing some statistical bias? Like, Carlsens loses are already pretty rare in classical, and by taking the game after a lose, you're by definition excluding one lose out of your sequence of results (Wins/Draws/Loses). So what remains of this sequence is then guaranteed to be of higher performance then the sequence you started with. I'm not sure how OP calculated it, (whether its just one game or just rest of tournament- i'm assuming the first one), but that would also mean it excludes prior results as well. for classical it would be biased towards prep work, and carlsen didn't really seem to care as much previously for blitz/rapid iirc, since his main time was going into prepwork for the classical world championships. he just happened to be a title holder of the other two.

@DaBassie said in #9:

For example, if I take some random short tournament sequence 110111011, my performance after a win is obviously much less than after a lose, which is in fact 100% here.

I am not sure about this. Perhaps the nulll hypothesis is that the sequence of loss and win is random, and unifrom over the event space of those 2 values. And then knowing that one event is either does not change the probability of the others. perhaps one have to consider varioius sizes of events. . proabbuility of ... but are they independent.

the null being that they would be independent (the question being are they not), and uniform etc.. (forgot all the words)..

then the questino might be whether we do have that null. and then some probability confidence looking at the result that it does not hold. I did not read the blog yet.. finnished the back ref one, as I enjoy the foundation work. did not have time today.. so I might be writing wind...

@DaBassie said in #9: > > For example, if I take some random short tournament sequence 110111011, my performance after a win is obviously much less than after a lose, which is in fact 100% here. I am not sure about this. Perhaps the nulll hypothesis is that the sequence of loss and win is random, and unifrom over the event space of those 2 values. And then knowing that one event is either does not change the probability of the others. perhaps one have to consider varioius sizes of events. . proabbuility of ... but are they independent. the null being that they would be independent (the question being are they not), and uniform etc.. (forgot all the words).. then the questino might be whether we do have that null. and then some probability confidence looking at the result that it does not hold. I did not read the blog yet.. finnished the back ref one, as I enjoy the foundation work. did not have time today.. so I might be writing wind...

I think these stats could use some peer review
Otherwise there maybe some hidden other variables affecting victory odds

Some obvious ideas that may or may not affect these results:

  1. Color. Win rate as black is usually smaller then win rate as white and in most tournaments you change your color (almost) every round. So you'll have a lot of white-black and black-white pairs of consecutive games and much smaller amount of white-white and black-black.
  2. Swiss pairing. Don't know how often Magnus plays swiss tournaments compared to full-round tournaments, but in swiss you'll be paired lower after a loss thus impacting your win chances
I think these stats could use some peer review Otherwise there maybe some hidden other variables affecting victory odds Some obvious ideas that may or may not affect these results: 1. Color. Win rate as black is usually smaller then win rate as white and in most tournaments you change your color (almost) every round. So you'll have a lot of white-black and black-white pairs of consecutive games and much smaller amount of white-white and black-black. 2. Swiss pairing. Don't know how often Magnus plays swiss tournaments compared to full-round tournaments, but in swiss you'll be paired lower after a loss thus impacting your win chances
<Comment deleted by user>

@Nikodimonius

Peer Review.

I second that but enlarge it to scientific spirit, discussion, research and idea exploration. Can be done right here, as you actually showed.

@Nikodimonius > Peer Review. I second that but enlarge it to scientific spirit, discussion, research and idea exploration. Can be done right here, as you actually showed.

I found the conclusions drawn from this statistical analysis to be quite interesting, whereas other analyses can be a bit dry. Good work, and please keep them coming!

I found the conclusions drawn from this statistical analysis to be quite interesting, whereas other analyses can be a bit dry. Good work, and please keep them coming!
<Comment deleted by user>

@DaBassie said in #9:

It's an interesting question for sure.

I was just wondering about your statistical method here. Isn't looking at 'the game after a lose' introducing some statistical bias? Like, Carlsens loses are already pretty rare in classical, and by taking the game after a lose, you're by definition excluding one lose out of your sequence of results (Wins/Draws/Loses). So what remains of this sequence is then guaranteed to be of higher performance then the sequence you started with.

For example, if I take some random short tournament sequence 110111011, my performance after a win is obviously much less than after a lose, which is in fact 100% here.

Might be negligible for high number of games though... I dunno. Interesting that rapid and blitz appear to be different, so perhaps even Carlsen is just tilting sometimes then.

That's an interesting point, I haven't thought about that. I'm not sure how to account for this, as the game performances aren't random. If the losses aren't randomly distributed but instead clumped together, the performance after a loss could also be much worse when there are only a few losses (in your example, you could have the two losses after each other, making the performance much worse).
If you have ideas to account for this, let me know

@DaBassie said in #9: > It's an interesting question for sure. > > I was just wondering about your statistical method here. Isn't looking at 'the game after a lose' introducing some statistical bias? Like, Carlsens loses are already pretty rare in classical, and by taking the game after a lose, you're by definition excluding one lose out of your sequence of results (Wins/Draws/Loses). So what remains of this sequence is then guaranteed to be of higher performance then the sequence you started with. > > For example, if I take some random short tournament sequence 110111011, my performance after a win is obviously much less than after a lose, which is in fact 100% here. > > Might be negligible for high number of games though... I dunno. Interesting that rapid and blitz appear to be different, so perhaps even Carlsen is just tilting sometimes then. That's an interesting point, I haven't thought about that. I'm not sure how to account for this, as the game performances aren't random. If the losses aren't randomly distributed but instead clumped together, the performance after a loss could also be much worse when there are only a few losses (in your example, you could have the two losses after each other, making the performance much worse). If you have ideas to account for this, let me know

@whatthefat said in #11:

Has colour been taken into account? If he is generally alternating colour he plays, and if losses are more likely as Black / wins are more likely as White, this alone could perhaps explain the trend in classical format.

I've checked the colours now, and in all time controls he has White in around 55% of the games after he lost.
In classical he scores 70% with White after a loss and 76% with Black after a loss. In rapid it is 68% and 62%, and in blitz it is 67% and 62%.
So there is a slight bias towards the white pieces after a loss, but in classical Carlsen scores better with black after a loss than with white.

@whatthefat said in #11: > Has colour been taken into account? If he is generally alternating colour he plays, and if losses are more likely as Black / wins are more likely as White, this alone could perhaps explain the trend in classical format. I've checked the colours now, and in all time controls he has White in around 55% of the games after he lost. In classical he scores 70% with White after a loss and 76% with Black after a loss. In rapid it is 68% and 62%, and in blitz it is 67% and 62%. So there is a slight bias towards the white pieces after a loss, but in classical Carlsen scores better with black after a loss than with white.