Carlsen withdrawal after loss to Niemann

[lkaufman]

This dropoff after round 3 is consistent with the huge difference in his performance in broadcast vs nonbroadcast events over a 2 year period reported in this thread. Until a couple days ago I thought that the cheating was confined to online, but now it doesn't look that way. Niemann is surely of grandmaster (FIDE 2500+) strength, but there is a huge difference between grandmaster strength and WC candidate (2800) strength, or even 2700 (FIDE) strength.

Is that not also maybe indicative of a player 'in over his head'...as it outclassed and simply falling back to Earth as the tourney wore on...and having to play consistently stronger players knowing each sitting across from him is wondering if he are cheating? That's some kind of pressure on Hans every game if you ask me...

That would have been my assumption prior to the data on his performance in broadcast vs nonbroadcast events, but that data is so convincing and so consistent with what happened in the current event that it is hard to ignore both, especially given the known online cheating.

[CornfedForever]

This dropoff after round 3 is consistent with the huge difference in his performance in broadcast vs nonbroadcast events over a 2 year period reported in this thread. Until a couple days ago I thought that the cheating was confined to online, but now it doesn't look that way. Niemann is surely of grandmaster (FIDE 2500+) strength, but there is a huge difference between grandmaster strength and WC candidate (2800) strength, or even 2700 (FIDE) strength.

Is that not also maybe indicative of a player 'in over his head'...as it outclassed and simply falling back to Earth as the tourney wore on...and having to play consistently stronger players knowing each sitting across from him is wondering if he are cheating? That's some kind of pressure on Hans every game if you ask me...

That would have been my assumption prior to the data on his performance in broadcast vs nonbroadcast events, but that data is so convincing and so consistent with what happened in the current event that it is hard to ignore both, especially given the known online cheating.

Someone said of the recent Miami Meltwater tourney that he thought the reason for the headphones was to mask any transmissions. I don't know if that was true - not sure why they would wear them, they may have just been 'noise cancelling'; but he thumped Magnus hard in the first game they played there...resulting in the now (in)famous quote as Magnus walked off behind him.

I'm sure the upcoming US Championship will take extra precautions.

[RandomGuy321]

Lame and missing the point. To demonstrate Hans Nieman’s data line is somehow special, it is necessary to compare it with data lines of all players (or a fairly selected subset of comparable players), otherwise known as a Control Group. Chess engine testers/rating list compilers will confirm that Elo is highly volatile, especially when based on small game samples (substitute humans for chess engines and you add a whole new layer of volatility). Delta-Elo (which the OP is using) even more so. Only a control group can show what is normal and what is an outlier. Without it no conclusions can even begin to be drawn. Basic statistics.
Take the OP data and table back to the OP with a big red line through it and a mark of 0/10. Wrong.

Once again the poster boy for Dunning-Kruger leaps at the chance to show he remains free of the trammels of knowledge, experience, and the capacity for learning and self-reflection!

You don't understand what a control group is. You apparently don't understand much of anything when it comes to the "basic statistics" you hold so dear. Observational studies don't have control groups. Instead, they use various methods to control for confounding variables.

Your incoherent criticism seems to be clumsily circling around the idea that you believe it would be useful to run a regression on data from more players, with added terms interacting each independent variable with another variable indicating whether a given player is Hans Niemann.

That's a reasonable thing to do, and you're welcome to do that here if you like. (Though at this point it's clear you wouldn't know how.) But I don't believe it's necessary in this instance, because:

• As I mentioned in my original analysis, and then again in later posts—you really must learn to read!—I already did this, albeit informally. I performed a cursory examination of a number of demographically similar players from the tournaments in the Niemann dataset to see if any of them displayed remotely similar patterns of performance. I found no one that did.

• AndrewGrant's devil's advocacy notwithstanding, it appears that nobody realistically believes there's a mechanism (other than cheating) that could explain a correlation between broadcast status and performance. In other words, there appears to be general consensus that the prior probability of a non-cheating mechanism is quite low. So we ignore it. This same reasoning is why, for example, I didn't entertain the idea that Niemann can read minds on Sundays.

• In effect, we ignore "possibilities" with evidently low prior probabilities, and are justified in doing so. This is common and reasonable in exploratory studies. You'll rarely be wrong to assume something is true when everyone believes it to be true, and nobody can even hypothesize a plausible alternate mechanism.

• By far the biggest question mark is the actual broadcast status of the tournaments in the Niemann dataset. The methodology of the analysis is completely standard, and sufficiently controlled for its purposes.

If Carlsen or Chess.com were to release a report from a team of investigators, I would demand that they be much more thorough than my exploratory observational study. I would want them to contact every tournament director, confirm broadcast status and manner of broadcast, gather data on broadcast delays and glitches (intentional or not), perform statistical analysis on the level of individual games, show distributions of various statistics over different populations of players by demographic, etc. In other words, I would want them to dot all the i's and cross all the t's. Only then would I be comfortable acting on the conclusions of the report, and allowing the chess community to formally sanction Niemann for cheating.

But for the purposes of figuring out whether it looks like there might be something here? To see if there's any evidence that Niemann might be or have been a cheater? That question has been answered definitively, from multiple independent angles. Yeah, there's evidence. There's a lot of it. It's time for people with resources to do a real investigation.

Hello, I have been following this conversation and was just curious if the sample size is a little small for the amount of variables you are analyzing in the regression. Also, is it really true that observational studies do not have control groups? https://projecteuclid.org/journals/stat ... 13232.full

[dkappe]

Hello, I have been following this conversation and was just curious if the sample size is a little small for the amount of variables you are analyzing in the regression. Also, is it really true that observational studies do not have control groups? https://projecteuclid.org/journals/stat ... 13232.full

Far too small. A more sensible approach would be to look at all OTB tournaments over the period in question and see where Hans lies relative to certain sub populations and individuals. Of those individuals who show the same marked difference in performance between broadcast and non broadcast tournaments, is there something they share in common (aside from cheating)?

[Chessqueen]

Hello, I have been following this conversation and was just curious if the sample size is a little small for the amount of variables you are analyzing in the regression. Also, is it really true that observational studies do not have control groups? https://projecteuclid.org/journals/stat ... 13232.full

Far too small. A more sensible approach would be to look at all OTB tournaments over the period in question and see where Hans lies relative to certain sub populations and individuals. Of those individuals who show the same marked difference in performance between broadcast and non broadcast tournaments, is there something they share in common (aside from cheating)?

Now Young super star GM Pragg has to start losing to Carlsen or Carlsen will accuse him of cheating too https://en.chessbase.com/post/generatio ... 22-preview


GM Ben Finegold Opinion

[DrCliche]

Hello, I have been following this conversation and was just curious if the sample size is a little small for the amount of variables you are analyzing in the regression.

No, not at all. You can rerun the regression with any subset of the explanatory variables (or just broadcast status alone) and get essentially the same results. All of the other regression coefficients were so insignificant (and there weren't that many, besides) that there's no particular reason to be concerned about overfitting or the curse of dimensionality.

Also, is it really true that observational studies do not have control groups? https://projecteuclid.org/journals/stat ... 13232.full

Yes, though terminology is sometimes abused, as you've seen. In observational studies, at best you can differentiate between a "treated" group and an "untreated" group, and then do your utmost to control for confounders. If you have no control over if, when, or how a treatment is assigned, you don't have a control group. (Moreover, absent random assignment, it's often ... challenging ... to reasonably claim that observed treatment effects are actually caused by the treatment. And even in so-called "natural experiments", where treatment exposure is alleged to simulate random assignment, you don't, strictly speaking, have a control group.)

Pedantry aside, chrisw was incorrectly using "control group" when he appeared to be suggesting that we shouldn't draw any conclusions without analyzing the statistics of Niemann's identically treated peers. The treatment in question is the broadcast status of various tournaments, which would be same for all players (ignoring the possibility of asymmetric broadcast glitches, or a tournament broadcasting only a subset of games, or whatever.)

What chrisw appeared to be proposing is really just sampling, and (more or less) looking where in the sample distribution Niemann falls. (For example, if we really did sample a bunch more players, and then performed the same regression analysis for each player, we might observe that Niemann's broadcast status regression coefficient was many standard deviations above the mean of the rest of the sample.) Assuming that's what chrisw meant, that's a reasonable next step to take, but it has nothing to do with a "control group".

[Chessqueen]

Hello, I have been following this conversation and was just curious if the sample size is a little small for the amount of variables you are analyzing in the regression.

No, not at all. You can rerun the regression with any subset of the explanatory variables (or just broadcast status alone) and get essentially the same results. All of the other regression coefficients were so insignificant (and there weren't that many, besides) that there's no particular reason to be concerned about overfitting or the curse of dimensionality.

Also, is it really true that observational studies do not have control groups? https://projecteuclid.org/journals/stat ... 13232.full

Yes, though terminology is sometimes abused, as you've seen. In observational studies, at best you can differentiate between a "treated" group and an "untreated" group, and then do your utmost to control for confounders. If you have no control over if, when, or how a treatment is assigned, you don't have a control group. (Moreover, absent random assignment, it's often ... challenging ... to reasonably claim that observed treatment effects are actually caused by the treatment. And even in so-called "natural experiments", where treatment exposure is alleged to simulate random assignment, you don't, strictly speaking, have a control group.)

Pedantry aside, chrisw was incorrectly using "control group" when he appeared to be suggesting that we shouldn't draw any conclusions without analyzing the statistics of Niemann's identically treated peers. The treatment in question is the broadcast status of various tournaments, which would be same for all players (ignoring the possibility of asymmetric broadcast glitches, or a tournament broadcasting only a subset of games, or whatever.)

What chrisw appeared to be proposing is really just sampling, and (more or less) looking where in the sample distribution Niemann falls. (For example, if we really did sample a bunch more players, and then performed the same regression analysis for each player, we might observe that Niemann's broadcast status regression coefficient was many standard deviations above the mean of the rest of the sample.) Assuming that's what chrisw meant, that's a reasonable next step to take, but it has nothing to do with a "control group".

Ben Finegold is correct, and all of you are misjudging GM Hans, simply because he is rated lower than Carlsen. Carlsen will need to apologize to GM Hans very soon. Carlsen believe that somebody from his team gave GM Hans his Opening Preparation, but if that is the case Carlsen could had opened with 1.a3 or 1.h3 as he has done before etc...
GM Ben Finegold Opinion

[RandomGuy321]

Hello, I have been following this conversation and was just curious if the sample size is a little small for the amount of variables you are analyzing in the regression.

No, not at all. You can rerun the regression with any subset of the explanatory variables (or just broadcast status alone) and get essentially the same results. All of the other regression coefficients were so insignificant (and there weren't that many, besides) that there's no particular reason to be concerned about overfitting or the curse of dimensionality.

Also, is it really true that observational studies do not have control groups? https://projecteuclid.org/journals/stat ... 13232.full

Yes, though terminology is sometimes abused, as you've seen. In observational studies, at best you can differentiate between a "treated" group and an "untreated" group, and then do your utmost to control for confounders. If you have no control over if, when, or how a treatment is assigned, you don't have a control group. (Moreover, absent random assignment, it's often ... challenging ... to reasonably claim that observed treatment effects are actually caused by the treatment. And even in so-called "natural experiments", where treatment exposure is alleged to simulate random assignment, you don't, strictly speaking, have a control group.)

Pedantry aside, chrisw was incorrectly using "control group" when he appeared to be suggesting that we shouldn't draw any conclusions without analyzing the statistics of Niemann's identically treated peers. The treatment in question is the broadcast status of various tournaments, which would be same for all players (ignoring the possibility of asymmetric broadcast glitches, or a tournament broadcasting only a subset of games, or whatever.)

What chrisw appeared to be proposing is really just sampling, and (more or less) looking where in the sample distribution Niemann falls. (For example, if we really did sample a bunch more players, and then performed the same regression analysis for each player, we might observe that Niemann's broadcast status regression coefficient was many standard deviations above the mean of the rest of the sample.) Assuming that's what chrisw meant, that's a reasonable next step to take, but it has nothing to do with a "control group".

Hey, thanks for getting back to me. I appreciate looking at the work you did since you sometimes don't get such interesting examples of statistical methods in a classroom setting (I took a class last semester). So, if he didn't cheat otb in this period, it would be historic right? I am curious how a predictive model would compare to Hans's performance this year if he were to stay on track from that specific date range you analyzed.

[Chessqueen]

Hello, I have been following this conversation and was just curious if the sample size is a little small for the amount of variables you are analyzing in the regression.

No, not at all. You can rerun the regression with any subset of the explanatory variables (or just broadcast status alone) and get essentially the same results. All of the other regression coefficients were so insignificant (and there weren't that many, besides) that there's no particular reason to be concerned about overfitting or the curse of dimensionality.

Also, is it really true that observational studies do not have control groups? https://projecteuclid.org/journals/stat ... 13232.full

Yes, though terminology is sometimes abused, as you've seen. In observational studies, at best you can differentiate between a "treated" group and an "untreated" group, and then do your utmost to control for confounders. If you have no control over if, when, or how a treatment is assigned, you don't have a control group. (Moreover, absent random assignment, it's often ... challenging ... to reasonably claim that observed treatment effects are actually caused by the treatment. And even in so-called "natural experiments", where treatment exposure is alleged to simulate random assignment, you don't, strictly speaking, have a control group.)

Pedantry aside, chrisw was incorrectly using "control group" when he appeared to be suggesting that we shouldn't draw any conclusions without analyzing the statistics of Niemann's identically treated peers. The treatment in question is the broadcast status of various tournaments, which would be same for all players (ignoring the possibility of asymmetric broadcast glitches, or a tournament broadcasting only a subset of games, or whatever.)

What chrisw appeared to be proposing is really just sampling, and (more or less) looking where in the sample distribution Niemann falls. (For example, if we really did sample a bunch more players, and then performed the same regression analysis for each player, we might observe that Niemann's broadcast status regression coefficient was many standard deviations above the mean of the rest of the sample.) Assuming that's what chrisw meant, that's a reasonable next step to take, but it has nothing to do with a "control group".

Hey, thanks for getting back to me. I appreciate looking at the work you did since you sometimes don't get such interesting examples of statistical methods in a classroom setting (I took a class last semester). So, if he didn't cheat otb in this period, it would be historic right? I am curious how a predictive model would compare to Hans's performance this year if he were to stay on track from that specific date range you analyzed.

Where is the evidence? ==>

[Chessqueen]

Hello, I have been following this conversation and was just curious if the sample size is a little small for the amount of variables you are analyzing in the regression.

No, not at all. You can rerun the regression with any subset of the explanatory variables (or just broadcast status alone) and get essentially the same results. All of the other regression coefficients were so insignificant (and there weren't that many, besides) that there's no particular reason to be concerned about overfitting or the curse of dimensionality.

Also, is it really true that observational studies do not have control groups? https://projecteuclid.org/journals/stat ... 13232.full

Yes, though terminology is sometimes abused, as you've seen. In observational studies, at best you can differentiate between a "treated" group and an "untreated" group, and then do your utmost to control for confounders. If you have no control over if, when, or how a treatment is assigned, you don't have a control group. (Moreover, absent random assignment, it's often ... challenging ... to reasonably claim that observed treatment effects are actually caused by the treatment. And even in so-called "natural experiments", where treatment exposure is alleged to simulate random assignment, you don't, strictly speaking, have a control group.)

Pedantry aside, chrisw was incorrectly using "control group" when he appeared to be suggesting that we shouldn't draw any conclusions without analyzing the statistics of Niemann's identically treated peers. The treatment in question is the broadcast status of various tournaments, which would be same for all players (ignoring the possibility of asymmetric broadcast glitches, or a tournament broadcasting only a subset of games, or whatever.)

What chrisw appeared to be proposing is really just sampling, and (more or less) looking where in the sample distribution Niemann falls. (For example, if we really did sample a bunch more players, and then performed the same regression analysis for each player, we might observe that Niemann's broadcast status regression coefficient was many standard deviations above the mean of the rest of the sample.) Assuming that's what chrisw meant, that's a reasonable next step to take, but it has nothing to do with a "control group".

Hey, thanks for getting back to me. I appreciate looking at the work you did since you sometimes don't get such interesting examples of statistical methods in a classroom setting (I took a class last semester). So, if he didn't cheat otb in this period, it would be historic right? I am curious how a predictive model would compare to Hans's performance this year if he were to stay on track from that specific date range you analyzed.

Where is the evidence? ==>

Top chess Players will have to play in Swimsuit or Naked surrounded with good looking girls Semi Naked, NOW that is what I call big entertainment Las Vegas Style ==>