[billw55]

campboy, on 2015-November-03, 04:03, said:

So, some proper figures. We have 65 hands. The experts pick 23 of these as "wide" hands. The video analysts pick 22 of them as "wide" hands.

Just to be clear, were they mostly the same hands? How many correlated?

[gwnn]

billw55, on 2015-November-03, 07:26, said:

Just to be clear, were they mostly the same hands? How many correlated?

38 out of the 40 where the experts were unanimous matched the signal. 18 out of the 25 where the experts were split (4:1 or 3:2, doesn't matter) also matched the signal (this makes for a total of 56/65). The 65 hands out of the original total 200 were the ones in which the wide/narrow question was unanimous from the observers.

edit: clarified/corrected the numbers.

This post has been edited by gwnn: 2015-November-03, 07:42

[campboy]

gwnn, on 2015-November-03, 07:07, said:

It's not difficult to inflate your figure? Could you have got such a figure for a random honest pair as well? I understand that 1 in 1B hypotheses will have a p-value of 10^(-9) for example, but do you really think we have that many (reasonably simple) hypotheses?

Of course it's trivial to inflate your figure. Just pick some thing that the pair does occasionally but not often, pick a piece of information that you'd want to show occasionally but not often, notice that on a lot of hands these things coincide (which they will, because most hands will be no for both), assume that both these things are actually 50% and there you go.

[billw55]

gwnn, on 2015-November-03, 07:37, said:

38 out of the 40 where the experts were unanimous matched the signal. 18 out of the 25 where the experts were split (4:1 or 3:2, doesn't matter) also matched the signal (this makes for a total of 56/65). The 65 hands out of the original total 200 were the ones in which the wide/narrow question was unanimous from the observers.

edit: clarified/corrected the numbers.

That sounds pretty convincing. Personally I would discard all the cases where the experts were split 3:2, but that is peanuts compared to the unanimous cases.

[Trinidad]

PhantomSac, on 2015-November-03, 04:13, said:

oh it's 1 in 17 million and not 1 in a billion OMG

The size of the numbers is important. Say that it would be the still large number of 1 in 100. That means that if you would video tape all the bridge hands that are played at the WC, you will find this good a correlation between the bidding card gap and the strength of the hand (in context) for several of the participants. That is because a large amount of hands are played.

So, the question is not "are the odds large?". The question is: "Are the odds much larger than the amount of times this could occur?". And now you are comparing two very large numbers to each other. This is something that is difficult to do without a good understanding of the underlying math.

If I spin a roulette wheel, it is only a very small probability that it comes up with '7' (1 in 37). If I put my money on '7' 3 times in a row and I win three times, people might look at me with some suspicion. "OMG! What are the odds?" (The answer is: 1 in 50 653). But I am sure that in a large casino this happens several times a year... because roulette wheels are spun over and over and over again... Many more times than 50 653.

This is a similar case. Bridge hands are played over and over and over again. That means that a number that seems very high may actually be small in comparison. And that means that it is important to know how high this number really is.

This all doesn't mean anything about B-Z's guilt or innocence. It means that the correct math is important.

Rik

[gwnn]

campboy, on 2015-November-03, 08:26, said:

Of course it's trivial to inflate your figure. Just pick some thing that the pair does occasionally but not often, pick a piece of information that you'd want to show occasionally but not often, notice that on a lot of hands these things coincide (which they will, because most hands will be no for both), assume that both these things are actually 50% and there you go.

It's trivial to inflate your figure but if someone says "whenever A-B scratch their nose after a truck passes by the building, they have either the diamond 7 or the club 9! that is true 38 out of 40 trials (total sample size:1 million)!!" it will be not as impressive as "whenever A-B make a wide bid, they have a non-minimum. it is true 38 out of 40 trials (total sample size: 200)" The first one is an awfully inflated figure for a trivial piece of information that nobody cares about (which was my point about the 1 billion things you would test for - most of the 1 billion will be useless trivia, not an interesting message to pass). The second one is a bit different.

[campboy]

gwnn, on 2015-November-03, 09:34, said:

It's trivial to inflate your figure but if someone says "whenever A-B scratch their nose after a truck passes by the building, they have either the diamond 7 or the club 9! that is true 38 out of 40 trials (total sample size:1 million)!!" it will be not as impressive as "whenever A-B make a wide bid, they have a non-minimum. it is true 38 out of 40 trials (total sample size: 200)" The first one is an awfully inflated figure for a trivial piece of information that nobody cares about (which was my point about the 1 billion things you would test for - most of the 1 billion will be useless trivia, not an interesting message to pass). The second one is a bit different.

I think you have got the wrong end of the stick. The linked article is not claiming that 38 out of 40 times when they made a wide bid they had a good hand. It is claiming that on 38 out of 40 hands they either made a wide bid and had a good hand, or didn't make a wide bid and didn't have a good hand. And that is why it becomes crucially important to find out how many of those 40 hands actually had a wide bid. If almost all of the 40 hands were bad hands where wide bids didn't occur, that proves nothing. As it happens, only 10 of those 40 hands had wide bids on them, and you actually have to look at the raw data and count stuff to get that information.

[gwnn]

OK sorry I just simplified my post for readability and it got sloppy. It should have been something like "make a wide/narrow bid, they have a maximum/minimum. (where unclear cases are deemed minimum and sometimes low ODR is also a minimum - whatever)" I agree that the 50% hypothesis is not a good approximation, but I still disagree with you that you can easily inflate a figure on any pair, if that is what you were claiming. See my post above on why. I don't think pointing out this sloppy phrasing invalidates my post.

This post has been edited by gwnn: 2015-November-03, 10:26

[campboy]

gwnn, on 2015-November-03, 10:24, said:

OK sorry I just simplified my post for readability and it got sloppy. It should have been something like "make a wide/narrow bid, they have a maximum/minimum. (where unclear cases are deemed minimum and sometimes low ODR is also a minimum - whatever)" I agree that the 50% hypothesis is not a good approximation, but I still disagree with you that you can easily inflate a figure on any pair, if that is what you were claiming. See my post above on why. I don't think pointing out this sloppy phrasing invalidates my post.

All you need to do is find an action that the player takes about 10% of the time, think of a hand feature you might be interested in signalling that comes up on about 10% of hands, record 65 hands, and get a little bit lucky. Then you assume 50% instead of 10% and get a figure of 1 in 1,000,000,000 for at least 56 out of 65 matches, when this actually has a probability of 1 in 4.

Now the wide bids in this article weren't anywhere near as rare as 10% (which is why you still get 17,000,000). But you actually need to go and count them to check that. They weren't anywhere near as common as 50% either.

[Vampyr]

nige1, on 2015-November-02, 10:36, said:

More convincing evidence about bidding card gaps from BridgeWinners

Kit and Co have done a great job but I still think it should be the WBF that instigaties and audits such investigations.

Indeed, and no one would agree with you more than "Kit and Co". In fact I seem to remember Boye expressing the wish that his efforts would kickstart the WBF's own procedures, if any. Instead they have released a statement saying, in effect, that they will tolerate cheating, but not investigation of cheating. They have vilified those who have volinteered loads of time and effort uncovering cheaters, rather than hiring them as investigators.

It really is time, as has been mentioned on BW, for another organisation to rise up and replace the WBF. It will be interesting, in any case, to see what happens in the next BB, when there will (I hope and expect) be a widespread boycott of the event. It would be great if a European NBO scheduled a major event at he same time. Maybe open only to qualifiers of BB, VC and the seniors thing.

[gwnn]

campboy, on 2015-November-03, 10:50, said:

All you need to do is find an action that the player takes about 10% of the time, think of a hand feature you might be interested in signalling that comes up on about 10% of hands, record 65 hands, and get a little bit lucky. Then you assume 50% instead of 10% and get a figure of 1 in 1,000,000,000 for at least 56 out of 65 matches, when this actually has a probability of 1 in 4.

Now the wide bids in this article weren't anywhere near as rare as 10% (which is why you still get 17,000,000). But you actually need to go and count them to check that. They weren't anywhere near as common as 50% either.

I understand this, but I still don't think you could (realistically) get such a high improbability for an innocent pair. For one, we are talking about making a signal vs not making one, and doing something that you can clearly see across the screen (say, making a face is irrelevant, adjusting your posture during bidding is irrelevant, etc). Also, for this inflation to work, you need low-frequency signals, which is the worst kind for encoding stuff. It's not a good idea to just signal whether my 1m is 18-19 balanced (I think the figure there is somewhere around 5-10%). It is actually a great idea to signal whether my Polish club 1M rebid is 12-14 or not 12-14 (the probability is actually quite close to 50% here, especially after the passes from opps). Finally, I feel obligated to point out that this 1 in 3 figure (22/65 right?) is not a good estimate for the "true" frequenc (I know you know this better than me but this is for the audience at home and I forgot my binomial distribution so it's for me too). The confidence interval for a 22 successes and 65 examples is 0.23..0.47 and assuming p=0.5, typical values for n (within two sigma) are 24..41 so 22 or 23 are not some epic huge discrepancy, although it is statistically significant. You can't just use the middle one for the "true" frequency. In fact, I would assume that the low end of that distribution is less likely than the high end due to encoding considerations. I am on board that the 1/1 billion is inflated. But I'm not on board with "trivially inflated for any pair", if that is what you were claiming.

[gwnn]

It would also be an interesting question how many times other pairs make clearly wide/narrow calls. For people who didn't read the methodology, Kit had a 5-point Likert scale (with a "couldn't see" option as a sixth choice) and required at least 2 out of 4 observers to deem a gap "clearly wide" and all 4 out of the 4 to deem it "somewhat/clearly wide." So even if one of the 4 observers say "couldn't see" or "normal," the whole observation is out. In these conditions, the "narrow" call is not just the absence of a signal, but the signal in and of itself. I don't think typical pairs would have 65/200 such calls, be they narrow or wide, but I know I am just conjecturing.

[Fluffy]

gwnn, on 2015-November-04, 02:42, said:

It would also be an interesting question how many times other pairs make clearly wide/narrow calls. For people who didn't read the methodology, Kit had a 5-point Likert scale (with a "couldn't see" option as a sixth choice) and required at least 2 out of 4 observers to deem a gap "clearly wide" and all 4 out of the 4 to deem it "somewhat/clearly wide." So even if one of the 4 observers say "couldn't see" or "normal," the whole observation is out. In these conditions, the "narrow" call is not just the absence of a signal, but the signal in and of itself. I don't think typical pairs would have 65/200 such calls, be they narrow or wide, but I know I am just conjecturing.

Yeah more like 3/200, and 2 of those 3 would be slam bids as final contracts.

Maybe I am exagerating as always, but I have rearranged bids from my opponents whose spacing I didn't like. I've payed "some" attention to that and only did it twice ever at top level (I do it constantly against LOLs).

[aguahombre]

Fluffy, on 2015-November-04, 04:48, said:

I have rearranged bids from my opponents whose spacing I didn't like. I've payed "some" attention to that and only did it twice ever at top level (I do it constantly against LOLs).

So, you are pretty much assuring us that a team of LOLs from Spain will not make it to the BB.

[Vampyr]

Fluffy, on 2015-November-04, 04:48, said:

Yeah more like 3/200, and 2 of those 3 would be slam bids as final contracts.

Maybe I am exagerating as always, but I have rearranged bids from my opponents whose spacing I didn't like. I've payed "some" attention to that and only did it twice ever at top level (I do it constantly against LOLs).

Are most events where you live played with screens?

[BillHiggin]

According to the "Neopolitan Club", the polish team is withdrawing from the upcoming EBL Championship Cup. http://neapolitanclu...mpions-cup.html
This was also reported on Bridge Winners but almost immediately erased.

[Trinidad]

BillHiggin, on 2015-November-04, 18:01, said:

According to the "Neopolitan Club", the polish team is withdrawing from the upcoming EBL Championship Cup. http://neapolitanclu...mpions-cup.html
This was also reported on Bridge Winners but almost immediately erased.

http://bridgewinners...s-b-z-analysis/

Rik

[Fluffy]

Vampyr, on 2015-November-04, 10:35, said:

Are most events where you live played with screens?

no

[Vampyr]

Fluffy, on 2015-November-07, 18:21, said:

no

So then you don't rearrange the bids of little old men and ladies "constantly".

[Vampyr]

Duplicate