Allegations that Darren Woods also known as Dooshcom on 888 poker is a cheater and a scammer
09-23-2011
, 02:46 AM
09-27-2011
, 01:17 AM
Quote:
Hello,
I posted earlier in the thread about the lack of actual cheating evidence in the OP, and suggested that if collusion occurred, it was probably by one colluder folding hands preflop when the other has a good hand.
Feruell contacted me and sent me an export of his database with the suspects' hands. I also sent an email to Dooshcom after he posted and offered to look at his hands in case he wanted to clear his name. He has not replied.
The Suspects
Dooshcom, benkaremail, liverfc123, benhamcheese are the suspects per OP. There is circumstantial evidence suggesting that the latter three are the same person, including benhamcheese showing up at 500-1000 5 mins after liverfc123 busted. I defer to Feruell on this evidence.
The analysis groups all four suspects and will be hereafter referred to as "Team". However, I would like to note that Dooshcom/benkaremail's hands are about 80% of the sample, so the statistical power of the following analysis applies mostly to those two players. I will follow up with another post looking at benhamcheese and liverfc123 individually. For now, I'll say they show the same pattern but on a smaller sample size.
The Analysis
I used the following approach in looking for evidence of collusion:
1) Filter for hands when a player has (77+, ATo+, ATs+, KQs) in the SB or BB and look at how often another player opens before him from the Button or Cutoff. The following splits were compared:
(a) Team player in BU/CO opening PFR* vs Team player in SB/BB
(b) Team player in BU/CO opening PFR* vs Everyone Else (not a Team member) in SB/BB
(c) Everyone Else in BU/CO opening PFR* vs Anyone in SB/BB . This last one is not necessary to reach a conclusion, but it's nice to have.
2) Filter for hands when a player saw Showdown in SB or BB. These are not all premium hands, but they are stronger than a random hand. Again, look at opening PFR* from another player in BU or CO.
- Same splits as above
* Important Technical Note: Filtering for SB/BB holecards biases the sample in a way that affects the observed BU/CO opening PFR. For this reason I am labeling the observed value PFR* to distinguish it from "true PFR". See note (i) below. That being said, we can still do apples-apples comparisons between the PFR* resulting from the 3 splits above.
RESULTS
The chart below shows the results for the case where the SB/BB has a premium hand. Notice that Team PFR* when another Team member holds the premium hand in the SB/BB is less than half of that when a non-Team member holds the premium hand. The results for the Button are statistically significant at the p < 0.0001 level. The Cutoff results are not statistically significant due to small sample size. (ii)
More Detailed Results
The following table shows the data used in the above chart, with sample sizes included. It has additional splits, Doosh vs Team and Team vs Doosh. This is to show that it didn't matter whether Doosh was the one opening the BU/CO or viceversa, they both have a low opening PFR* when the other has a good hand in the blinds.
It also shows an "Inferred PFR" - This is an estimate of the true PFR from PFR*, which depends on the opening/calling frequencies of the players that act after the BU/CO. I used the regulars' frequencies for the purpose of this estimate. It gives an idea of the players' actual PFR. Note that we estimate that the Team was opening < 20% from the button when their buddy had a premium hand in the BB. Their overall button opening PFR was 46%.
Compare the "Team vs Team" and "Team vs Everyone Else" rows in the table below. The p value refers to a statistical comparison between those two rows.
Lastly, below are the results when the buddy had any hand that went to showdown in the SB/BB. It still shows the same pattern, statistically significant for all cases except CO vs BB which has too few samples. This suggests that the collusion was not limited to the most premium hands.
Note that estimating the true PFR in this case is more complicated. The inferred PFR is not always smaller than PFR*, as was the case when we filtered by premium hands. If anyone's interested in the technical details, let me know.
That's it for now, I think this is conclusive, more to come later.
CLIFFS: Dooshcom and team were opening about half the normal amount when a buddy had a premium hand in the blinds.
(i) The easiest way to see this is to imagine that the BB has a premium hand, and will always play as a result. When the BU/CO raise all-in, the BB will call and all those hands are in the sample. Likewise when the BU/CO fold but another player puts BB all-in. However, when all players fold to the BB, the hand doesn't go to showdown and is not included in the filtered sample. Thus the sample PFR* in this case will be higher than the true PFR since it doesn't include those times when BU/CO folded and the hand didn't go to showdown.
(ii) I'm not a stastician, but I googled around and used the tests on the following web page, which seemed to be the right tests to use. The proportions compared were the Team's PFR* when a buddy had a good hand in the SB/BB vs the same stat when the good hand is held by someone else.
http://faculty.vassar.edu/lowry/tab2x2.html
I posted earlier in the thread about the lack of actual cheating evidence in the OP, and suggested that if collusion occurred, it was probably by one colluder folding hands preflop when the other has a good hand.
Feruell contacted me and sent me an export of his database with the suspects' hands. I also sent an email to Dooshcom after he posted and offered to look at his hands in case he wanted to clear his name. He has not replied.
The Suspects
Dooshcom, benkaremail, liverfc123, benhamcheese are the suspects per OP. There is circumstantial evidence suggesting that the latter three are the same person, including benhamcheese showing up at 500-1000 5 mins after liverfc123 busted. I defer to Feruell on this evidence.
The analysis groups all four suspects and will be hereafter referred to as "Team". However, I would like to note that Dooshcom/benkaremail's hands are about 80% of the sample, so the statistical power of the following analysis applies mostly to those two players. I will follow up with another post looking at benhamcheese and liverfc123 individually. For now, I'll say they show the same pattern but on a smaller sample size.
The Analysis
I used the following approach in looking for evidence of collusion:
1) Filter for hands when a player has (77+, ATo+, ATs+, KQs) in the SB or BB and look at how often another player opens before him from the Button or Cutoff. The following splits were compared:
(a) Team player in BU/CO opening PFR* vs Team player in SB/BB
(b) Team player in BU/CO opening PFR* vs Everyone Else (not a Team member) in SB/BB
(c) Everyone Else in BU/CO opening PFR* vs Anyone in SB/BB . This last one is not necessary to reach a conclusion, but it's nice to have.
2) Filter for hands when a player saw Showdown in SB or BB. These are not all premium hands, but they are stronger than a random hand. Again, look at opening PFR* from another player in BU or CO.
- Same splits as above
* Important Technical Note: Filtering for SB/BB holecards biases the sample in a way that affects the observed BU/CO opening PFR. For this reason I am labeling the observed value PFR* to distinguish it from "true PFR". See note (i) below. That being said, we can still do apples-apples comparisons between the PFR* resulting from the 3 splits above.
RESULTS
The chart below shows the results for the case where the SB/BB has a premium hand. Notice that Team PFR* when another Team member holds the premium hand in the SB/BB is less than half of that when a non-Team member holds the premium hand. The results for the Button are statistically significant at the p < 0.0001 level. The Cutoff results are not statistically significant due to small sample size. (ii)
More Detailed Results
The following table shows the data used in the above chart, with sample sizes included. It has additional splits, Doosh vs Team and Team vs Doosh. This is to show that it didn't matter whether Doosh was the one opening the BU/CO or viceversa, they both have a low opening PFR* when the other has a good hand in the blinds.
It also shows an "Inferred PFR" - This is an estimate of the true PFR from PFR*, which depends on the opening/calling frequencies of the players that act after the BU/CO. I used the regulars' frequencies for the purpose of this estimate. It gives an idea of the players' actual PFR. Note that we estimate that the Team was opening < 20% from the button when their buddy had a premium hand in the BB. Their overall button opening PFR was 46%.
Compare the "Team vs Team" and "Team vs Everyone Else" rows in the table below. The p value refers to a statistical comparison between those two rows.
Lastly, below are the results when the buddy had any hand that went to showdown in the SB/BB. It still shows the same pattern, statistically significant for all cases except CO vs BB which has too few samples. This suggests that the collusion was not limited to the most premium hands.
Note that estimating the true PFR in this case is more complicated. The inferred PFR is not always smaller than PFR*, as was the case when we filtered by premium hands. If anyone's interested in the technical details, let me know.
That's it for now, I think this is conclusive, more to come later.
CLIFFS: Dooshcom and team were opening about half the normal amount when a buddy had a premium hand in the blinds.
(i) The easiest way to see this is to imagine that the BB has a premium hand, and will always play as a result. When the BU/CO raise all-in, the BB will call and all those hands are in the sample. Likewise when the BU/CO fold but another player puts BB all-in. However, when all players fold to the BB, the hand doesn't go to showdown and is not included in the filtered sample. Thus the sample PFR* in this case will be higher than the true PFR since it doesn't include those times when BU/CO folded and the hand didn't go to showdown.
(ii) I'm not a stastician, but I googled around and used the tests on the following web page, which seemed to be the right tests to use. The proportions compared were the Team's PFR* when a buddy had a good hand in the SB/BB vs the same stat when the good hand is held by someone else.
http://faculty.vassar.edu/lowry/tab2x2.html
Also saying since "statistics" proves that they did cheat it must be true isn't a good argument. If they do poor statistics, incorrect statistics, or don't know how to do statistics then all they really prove is they suck at math.
I just want to bang my head against the wall. Your first post is (a) very unconvincing (b) clearly well refuted by the guy being accused, and then you choose to (c) post NEW evidence that COULD be correct. Christ do you just get unlimited shots at proving a guy is a scammer?
I seriously think something has to be said for providing a coherent and convincing argument to begin with. Starting off weak as hell, and then trying to improve is ridiculous. Now you're just hoping he says something "off" so that you can use that as evidence.
Sigh, seriously I think you fail at this one. Obviously I feel no need to go figure out what the proper statistics is for this scenario. I would rather wait and do statistics on stuff that is pertinent to my life.
09-27-2011
, 01:34 PM
Quote:
Alright, so you're anology doesn't fully fit. I agree you CAN prove with a sample that something is highly unlikely to be true, or likely to be true. However, I'm not convinced he did the math/statistics part of the problem correctly. From an outside glance they look incorrect. What you are missing is VARIANCE. The variance of a 20 hand sample will be larger than a 50 hand sample will be larger than a 100 hand sample etc. etc. So directly comparing the means of two samples that are fundamentally different in size, and thus variance, is not valid.
It's like saying because I sat down at a poker table today, played 100 hands, and made 100 bucks. Then I sat down at a poker table tomorrow, played 300 hands, and only made 20 bucks. Then I compare these two samples and conclude that it is highly unlikely I should have expected to only make 20 bucks the second time. I don't see how what he's doing is much different from that (literally, he's using samples that are very comparative for each positional play to what I'm using in my example, and making similar conclusions). You also cannot say that because everyone else made either 50 bucks, or lost 50 bucks, that I was cheated. That doesn't work either.
It's like saying because I sat down at a poker table today, played 100 hands, and made 100 bucks. Then I sat down at a poker table tomorrow, played 300 hands, and only made 20 bucks. Then I compare these two samples and conclude that it is highly unlikely I should have expected to only make 20 bucks the second time. I don't see how what he's doing is much different from that (literally, he's using samples that are very comparative for each positional play to what I'm using in my example, and making similar conclusions). You also cannot say that because everyone else made either 50 bucks, or lost 50 bucks, that I was cheated. That doesn't work either.
However, your criticism is uninformed and misguided. Your concerns about how much variation in outcomes can be expected over a given sample are addressed in the analysis. Jaytorr concludes:
Quote:
RESULTS
The chart below shows the results for the case where the SB/BB has a premium hand. Notice that Team PFR* when another Team member holds the premium hand in the SB/BB is less than half of that when a non-Team member holds the premium hand. The results for the Button are statistically significant at the p < 0.0001 level. The Cutoff results are not statistically significant due to small sample size. (ii)
The chart below shows the results for the case where the SB/BB has a premium hand. Notice that Team PFR* when another Team member holds the premium hand in the SB/BB is less than half of that when a non-Team member holds the premium hand. The results for the Button are statistically significant at the p < 0.0001 level. The Cutoff results are not statistically significant due to small sample size. (ii)
Here is how measures of statistical significance work: http://en.wikipedia.org/wiki/Statistical_significance
Again, I'm not enough of an expert and haven't looked at the data closely enough to be certain he's doing everything right, but the problems you're pointing out are not the issue. If he's making a mistake it is a more subtle one than "failing to account for variance," and nothing like making a conclusion about a cash game winrate based on 100 hands.
