Right around 1400 there is a weird borderline. First the joy of passing it, then sliding beneath it. I also wondered how I could lose several in a row, when just passing 1400. There might be similar around other levels too, but breaking 1400 is "finally I know something about chess!"
Humans tend to be notoriously bad at generating random numbers. Indeed en.wikipedia.org/wiki/Benford%27s_law is used in fraud detection for this very reason.
As far as streaks go part of the reason for the higher incidence is that humans don't have a fixed Elo. For example a player that is stressed, tired and distracted will play a lot worse than same player that is relaxed, well rested and fully focused on the game. So this means in a hypothetical 1400 vs 1400 scenario the stressed scenario is a 1200 vs 1400 in reality while the relaxed scenario could be a 1600 vs 1400 scenario. This can equally apply to opponents so gap can be narrowed or widened.
If playing opponents that are tilting then you can potentially even lock in a 1600 vs 1200 scenario that can generate big streaks.
As far as streaks go part of the reason for the higher incidence is that humans don't have a fixed Elo. For example a player that is stressed, tired and distracted will play a lot worse than same player that is relaxed, well rested and fully focused on the game. So this means in a hypothetical 1400 vs 1400 scenario the stressed scenario is a 1200 vs 1400 in reality while the relaxed scenario could be a 1600 vs 1400 scenario. This can equally apply to opponents so gap can be narrowed or widened.
If playing opponents that are tilting then you can potentially even lock in a 1600 vs 1200 scenario that can generate big streaks.
Why does Chess.com have a "Losing-streak" ?
@KMcGeoch said in #3:
> Humans tend to be notoriously bad at generating random numbers. Indeed en.wikipedia.org/wiki/Benford%27s_law is used in fraud detection for this very reason.
>
> As far as streaks go part of the reason for the higher incidence is that humans don't have a fixed Elo. For example a player that is stressed, tired and distracted will play a lot worse than same player that is relaxed, well rested and fully focused on the game. So this means in a hypothetical 1400 vs 1400 scenario the stressed scenario is a 1200 vs 1400 in reality while the relaxed scenario could be a 1600 vs 1400 scenario. This can equally apply to opponents so gap can be narrowed or widened.
>
> If playing opponents that are tilting then you can potentially even lock in a 1600 vs 1200 scenario that can generate big streaks
Your idea of something like "effective ELO" is interesting, but the challenge is how to measure it. I agree that fluctuations in stress levels, etc. are probably a contributor to these streaks, but whether there is a way to turn that into something quantitative that could be used to expand the null model in the paper is very much an open question! Do you have sources for your 1200/1600 range or is that your intuition? I'd be very interested to read something estimating the ELO impact of different moods/physiological states.
> Humans tend to be notoriously bad at generating random numbers. Indeed en.wikipedia.org/wiki/Benford%27s_law is used in fraud detection for this very reason.
>
> As far as streaks go part of the reason for the higher incidence is that humans don't have a fixed Elo. For example a player that is stressed, tired and distracted will play a lot worse than same player that is relaxed, well rested and fully focused on the game. So this means in a hypothetical 1400 vs 1400 scenario the stressed scenario is a 1200 vs 1400 in reality while the relaxed scenario could be a 1600 vs 1400 scenario. This can equally apply to opponents so gap can be narrowed or widened.
>
> If playing opponents that are tilting then you can potentially even lock in a 1600 vs 1200 scenario that can generate big streaks
Your idea of something like "effective ELO" is interesting, but the challenge is how to measure it. I agree that fluctuations in stress levels, etc. are probably a contributor to these streaks, but whether there is a way to turn that into something quantitative that could be used to expand the null model in the paper is very much an open question! Do you have sources for your 1200/1600 range or is that your intuition? I'd be very interested to read something estimating the ELO impact of different moods/physiological states.
@Rockiron said in #2:
> Right around 1400 there is a weird borderline. First the joy of passing it, then sliding beneath it. I also wondered how I could lose several in a row, when just passing 1400. There might be similar around other levels too, but breaking 1400 is "finally I know something about chess!"
Your comment reminds me of the saw-tooth edges you can see in the Lichess rating distributions - anxiety about crossing a century threshold in either direction may be a specific contributor to streak-y play! It would be neat to look at how streaks are affected by playing in Zen mode or otherwise not knowing your rating or the opponent's rating, or how streakiness is affected by proximity to landmark ratings. Thanks for reading!
> Right around 1400 there is a weird borderline. First the joy of passing it, then sliding beneath it. I also wondered how I could lose several in a row, when just passing 1400. There might be similar around other levels too, but breaking 1400 is "finally I know something about chess!"
Your comment reminds me of the saw-tooth edges you can see in the Lichess rating distributions - anxiety about crossing a century threshold in either direction may be a specific contributor to streak-y play! It would be neat to look at how streaks are affected by playing in Zen mode or otherwise not knowing your rating or the opponent's rating, or how streakiness is affected by proximity to landmark ratings. Thanks for reading!
I think it's close to impossible to measure. Since the only way to measure it is through actual performance. The problem is that if estimating how likely a streak is based on ratings then we're effectively doing streak calculation in reverse to get a rating that explains the streak.
Probably the only realistic thing to do is to get Tournament Performance Ratings for people with relatively stable ratings and see how far it deviates from their rating. That probably gives a relatively good idea of upper and lower bounds. However where people lie within them is very hard to tell. It can even change on a game by game basis as in the morning I could feel fresh and after a long morning game I could feel worn out for the afternoon game.
If really wanting an experimental model then one method could be to get the respondents to first do a puzzle racer or puzzle storm exercise as per their preference and then play the game. The puzzle racer/storm exercise is then collated along with results. I would expect that when they perform above average on puzzle game they'll also perform above average in game. In same way a below average performance could translate in a similar way to the game
As far as to how 1200 and 1600 figures were taken it's merely figures plucked out of thin air with no basis behind them. However if they are valid then a +400 gap has an expected score of 0.9091 as opposed to 0.5000 that would have a big impact on streaks.
I'm not too sure there's much literature with respect to ELO impact of different moods/physiological states. I remember reading Practical Chess Psychology by Amatzia Avni although that dealt more with player's mindsets than how performance was affected by it. It might be more domain of sports psychologists since presumably research in other sports can be applied to chess
Probably the only realistic thing to do is to get Tournament Performance Ratings for people with relatively stable ratings and see how far it deviates from their rating. That probably gives a relatively good idea of upper and lower bounds. However where people lie within them is very hard to tell. It can even change on a game by game basis as in the morning I could feel fresh and after a long morning game I could feel worn out for the afternoon game.
If really wanting an experimental model then one method could be to get the respondents to first do a puzzle racer or puzzle storm exercise as per their preference and then play the game. The puzzle racer/storm exercise is then collated along with results. I would expect that when they perform above average on puzzle game they'll also perform above average in game. In same way a below average performance could translate in a similar way to the game
As far as to how 1200 and 1600 figures were taken it's merely figures plucked out of thin air with no basis behind them. However if they are valid then a +400 gap has an expected score of 0.9091 as opposed to 0.5000 that would have a big impact on streaks.
I'm not too sure there's much literature with respect to ELO impact of different moods/physiological states. I remember reading Practical Chess Psychology by Amatzia Avni although that dealt more with player's mindsets than how performance was affected by it. It might be more domain of sports psychologists since presumably research in other sports can be applied to chess
IMHO it's quite natural that stronger players are less prone to streaks. I believe it's a necessary part of improvement to learn to focus on the game fully and not let other factors (like having lost six games in a row) affect your play. Only statistically, of course, I'm quite sure there are also big differences even between players of similar level.
My experience also confirms that longer temporal gaps between games do not really help. Playing 2-3 games per week doesn't really make the fact that you lost six games in a row less frustrating or easier to ignore. Sure, playing another game right after a demoralizing loss is tough but if it's few days later, you had time analyze the game, think about it and about what you could (and should) have done, more time for anxiety if you are able to "redeem yourself" etc.
My experience also confirms that longer temporal gaps between games do not really help. Playing 2-3 games per week doesn't really make the fact that you lost six games in a row less frustrating or easier to ignore. Sure, playing another game right after a demoralizing loss is tough but if it's few days later, you had time analyze the game, think about it and about what you could (and should) have done, more time for anxiety if you are able to "redeem yourself" etc.
About 30 years ago, I had a winless OTB streak of 22 games. I think it was something like +0 =19 -3. I lost one game on time on move 40 just as I was making my move while up a piece. In another I blundered up 2Ps in a minor piece ending and only drew.
I was already an IM when this happened. Bad luck creates bad play which creates more bad luck. My suggestion is to relax and realize that what doesn't kill you makes you mentally stronger ... hopefully.
I was already an IM when this happened. Bad luck creates bad play which creates more bad luck. My suggestion is to relax and realize that what doesn't kill you makes you mentally stronger ... hopefully.
I think the higher streakiness between beginners is pretty interesting. I'd hypothesize a simple explanation. Ratings reflect relative strength, but not objective strength. In practice this results in a system where real strength differences are greater at higher ratings. This makes perfect sense if you view playing strength as distributed on something like a power law curve.
What this means is that the difference in objective playing strength between a 2400 and 2800 is far greater than that between a 1000 and a 1400, even though it's just 400 relative points in both cases. So if running hot can add a relatively comparable amount of 'objective' strength to both beginners and strong players, it'd have a much larger influence on the results of beginners, because it'd be a much larger increase in relative strength (or ELO) than it would be for a strong player.
The only way this wouldn't be true is if running hot added a much higher amount of 'objective strength' to stronger players, and I see no logical reason to think that even *might* be true.
What this means is that the difference in objective playing strength between a 2400 and 2800 is far greater than that between a 1000 and a 1400, even though it's just 400 relative points in both cases. So if running hot can add a relatively comparable amount of 'objective' strength to both beginners and strong players, it'd have a much larger influence on the results of beginners, because it'd be a much larger increase in relative strength (or ELO) than it would be for a strong player.
The only way this wouldn't be true is if running hot added a much higher amount of 'objective strength' to stronger players, and I see no logical reason to think that even *might* be true.