Fascinating analysis. This certainly shuts down many of the proposed alternate explanations for this result. The only confounding explanation I could think of for this result is that Hans performs significantly better when under media glare. (This would result in the dependent variable correlating with the error term in your regression, a potential endogeneity concern when you make the leap to causation). While this is probably true for Hans, I dont think anyone realistically believes it would make THAT much of a difference in performance. A difference in performance of 100+ ELO is ridiculously high to be explained by such a hypothesis.DrCliche wrote: ↑Sun Sep 11, 2022 6:20 am Here's some more fuel for the fire, but the tl;dr is that Niemann is overwhelmingly likely to be a chronic and habitual OTB cheater. Moreover, he was likely fiddling with a device behind his right ear in one of his Sinquefield interviews.
Without further ado:
----
https://twitter.com/atl_kings/status/1568656197812891653
The data is a contiguous block of classical tournaments in which Niemann played >0 games against opposition rated >2000 USCF on average.
For some of the listed tournaments I was unable to definitively confirm whether or not games were broadcast live in 2019/2020, but in zero instances did I find a tournament website that contradicted the claims of the above table. For the rest of my analysis I assume that the broadcast status claimed in the tweet is correct. I was unable to find any mentions of broadcast delay, and for the rest of my analysis I assume there was none.
If you run a regression on the above dataset, you will find that whether or not a tournament was broadcast live explains 67% (!!!!!) of the variation in Niemann's performance over that time period. (The rest of the variation is likely random, or at least isn't explained by age, number of rounds, or strength of opponents.)
Here are the regression coefficients, their p-values, and confidence intervals:
As you can see, the coefficients for time (measured in months), number of rounds, and average opponent all have very high p-values, and are statistically indistinguishable from 0.
The way to interpret the coefficients is as follows:
This is incredibly damning, and in my mind confirms beyond a shadow of a doubt that Niemann was cheating OVER THE BOARD at every available opportunity in 2019 and 2020. There's no reason to believe that he didn't and doesn't simply continue to cheat over the board at every available opportunity, period.
- Time — for each month that has passed since March 2019, Niemann is expected to gain 0.21 more Elo at a given tournament, though 0 is in the 95% confidence interval. (i.e. Niemann doesn't appear to be improving over time.)
- LiveCast — if there's a live broadcast at a given tournament, Niemann is expected to gain 23.11 more Elo at that tournament, with a p-value of 0.0009, and a 95% CI of [11, 35]. (Holy shit!)
- NumRounds — for each game played at a given tournament, Niemann is expected to lose 0.78 more Elo at that tournament, though 0 is in the 95% confidence interval. (i.e. Niemann doesn't appear to be affected by fatigue.)
- AvgOpp — if you increase the strength of his average opponent at a given tournament by 100 Elo, Niemann is expected to gain 2 more Elo at that tournament, though 0 is in the 95% confidence interval. (i.e. Niemann doesn't appear to be "playing up" or "playing down" to his competition, nor to be affected by mathematical caps on his performance rating against low rated opponents.)
----
The complaints regarding the lack of control, as mentioned by some on this forum, are frankly ridiculous.
However, the biggest concern here is the accuracy of the Live / Not Live classification.
1. AtlantaChess (original tweeter) gives no source for the classification.
2. Strong Chess, replied that 5 tournaments were misclassified (US Masters /Philly Intl / World Open etc). He also gives no source for his claim, other than suggesting that he was 'present' for (at least) one of the tournaments.
3. There is a discussion on only 1 of the 5 'misclassified' tournaments - the US Masters. StrongChess refutes AtlantaChess by saying that 'I was there. the projector only had a couple top boards, and most boards were not on the DGTs. The TDs manually transcribed scoresheets into the DB after the rounds. I even have a memory of Hans sitting in the non-DGT area playing a lower rated one round.' AtlantaChess does not contradict this, nor does he challenge the misclassification allegation for the other 4 tournaments.
You yourself showed that the conclusions of your analysis fall below significance if 2+ (relevant) tournaments are classified wrong. Since this is a question of 5 tournaments, the validity of your analysis rests on whether you believe AtlantaChess or StrongChess. And based on point 3 above, I lean towards Strong Chess.
