Quote:
Originally Posted by Donger Kim
Also, 15k hands is a very solid sample in HUNL
A solid sample for what? Determining who is the better player?
If one player is up 10bb/100 after the 15K hands we would declare him the winner and better player right? Is this statistically valid? If we assume a 135bb/100 standard deviation we cannot even say with 70% confidence that the player who is up 10bb/100 after 15K hands is even a winning player. If we assume the player is up an absurd 20bb/100 after 15K hands then we cannot say with 95% confidence that he is even a winning player!
While I may be mistaken about this, the win rates in the previous examples were quite high considering it is two top players. If we assume that one player has a more modest 5bb/100 win rate versus the other player. The player with 5bb edge is going to be down money at the end of the 15K sample 33% of the time. At a 3bb/100 edge for one player, said player has a 39% chance of losing over the 15K sample. Even over 100K hands a 3bb winner with 135bb standard deviation will lose 24% of the time.
These challenges are fun to watch and people like them. I don't claim to know who is the better player and frankly do not care. With that said, it is quite silly to think that if the players are even reasonably matched that these challenges actually decide who the better player is. The fact of the matter is that as relative skill is "close" randomness is the primary factor in who is up or down at the end of the sample.