Quote:
Originally Posted by Kelvis
I still don't see what work you've done on calculating the correlation or significance.
What are you going to do with a bigger sample other than give anecdotal evidence? I want to see what method you have used on this tiny sample that you intend to use on a bigger one. Do you have one?
Kelvis, you're right in that it mostly appears to be anecdotal evidence. In assessing this small sample, I am looking at instances where winning hole cards are dealt to a player within 1-4 hands of them losing a significant number of chips (more than just the ante, but blinds do get counted as losses). By "winning hole cards", I mean those hole cards that would be the "solution" to the current board if it were played out to the river. (Sometimes we know what the winning hole cards are before the river hits, so this does allow for additional data points.)
The expected incidence of the above instances will vary depending on the number of players at the table. I propose that the above should happen for a:
9 player table - 11% of the time (winning hole cards on the next hand), 20% of the time (by the second hand), 30% of the time (by the third hand), and 38% of the time (by the fourth hand).
8 player table - 13% of the time (next hand), 23% of the time (by the second hand), etc. etc.
For the February 11th, 2018 sample set, there are ~170 hands. Out of those hands, the vast majority were folded so it is impossible to tell what hole cards would have won for sure. Out of ~170 hands, there are 48 hands where it is possible to see whose hole hand cards would win (or if it was a tie).
Out of the 48 hands, 12 were on a 9 player table.
Of those 12 hands that could be analyzed, 5 (41%) would have been won by the player that had lost the most chips in the previous hand and while 3 (25%) would have been won by the player that had lost the most chips two hands ago. That makes 8 (66%) of won hands by "net chip losers" from the last one or last two hands. The other 4 (33%) winners did not appear to follow any pattern on first pass, but may also be due to the influence of folded winning hole cards in previous hands.
Out of the remaining hands, only 1 full hand played on a 8 player table. The player that got busted in that hand was busted on a board that should have been won by a player who lost the most chips in the previous two hands, but instead he got busted by the player who lost the most chips in the previous hand (with Aces).
Out of the remaining hands, only 1 full hand played on a 7 player table. The player that got busted in that hand should have been busted by the player who lost the most chips over the previous two hands, but instead was busted by the player who lost the most chips in the previous hand (with a set of 3s).
Out of the remaining hands, only 1 full hand played on a 6 player table. The player that got busted in that hand got busted by the player who lost the most chips in the previous hand. The true winner (who had folded) was a less active player who had also been losing chips over the last few hands.
Out of the remaining hands, only 1 full hand played on a 5 player table. A player got busted by a net chip loser from the previous hand.
Out of the remaining hands, only 1 full hand played on a 4 player table. The net winner from the previous hand got busted by the net chip loser from the past two hands.
The pattern persists further, as the remaining busts were also on net winners in previous hands by net losers in previous hands. However there is no useful statistical analysis that can be applied with these few hands when considering play on such small table sizes.
10 of 48 hands were three-way - 7 were (or should have been) won by net chip losers from previous hand, 3 appeared random. (70%/30%)
21 of the 48 hands were heads up and there is no obvious pattern. The amount of folding completely washes out the data.
I am having trouble designing a statistical test that would capture all of this information fairly, all at once. The expected percentages will change depending on the number of players at a table, so it looks like I would have to partition the audit and look at hands with constant table sizes.
I am thinking of dividing the data points for each table size into a winning group and a losing group. I will further partition each group:
Losers that lost to a net chip winner (counted over the past 1-2 hands) who had the winning hole cards to the board and losers that lost (or should have lost) to a net chip loser (counted over the past 1-2 hands, including loss of blinds but excluding antes for now) who had the winning hole cards to the board.
Likewise, the winners will be partitioned in a similar fashion except that their victims, the losing players, will be partitioned based on whether they were net chip winners over the past 1-2 hands or net chip losers.
I need to figure out what statistical test to use to determine correlation and significance for this analysis. I am thinking to compare the occurrence rates of each type of situation in each group (ex. winning or losing under those specific conditions) versus expected baselines (ex. 88% to lose at a 9 player table or 11% to win at a 9 player table). Then I would compare the occurrence rates to determine significance? How does that sound, Kelvis? Dog's breakfast? There might be a simpler way to go about it.
Last edited by puzzlefish; 02-20-2018 at 04:02 PM.