Everyone sits South in BBO

[pilowsky]

Like a bad conspiracy theory, or a piece of Soviet-era propaganda, Some questions come up very frequently on this Forum, so I created a small experiment to answer some of them.

I set up an IMPS table in the Prime area with a robot at each seat. I then let the robots play 6 sets of 12 hands - 72 boards (see Figure).

Each hand is also played elsewhere in BBO world against robots and with a human in one seat. The human is generally sitting South.

Here are the results:

 

 

Is there bias in the way that BBO deals the hands?

If this were the case then after playing 6 sets of boards one of NS or EW would have a significantly higher number of IMPS. Chi-square test is negative. The sum of wins and losses is equal to 0 therefore no bias exists.

 

Which is the best seat?

There is no bias so no seat is better than any other, but I recommend that you wear a mask and avoid television.

 

Which robot is the best partner?

This study was conducted using advanced robots. It would be interesting to see how much quantitative difference there is when pitting two advanced robots against two basic robots, or playing a 'team of three'.

 

Other conclusions.

There is a bias towards South in the Bridge community for historical reasons [1] and most players in the Prime area choose to sit South. BBO perpetuates this by seating players South in 'Best Hand' tournaments. Perhaps players in Prime and elsewhere jump for the South seat in the expectation that there are hands floating about in the ether cast adrift from 'Best-Hand' tournaments, and that if they sit South they will benefit from this. Like panning for gold in the effluent from a gold mine.

This seems not to be the case.

I still like to sit South though.

Of course, it doesn't help that whichever seat you choose, you are always at the bottom of the screen. Coming from the Southern hemisphere I'm used to it, but some in the North, East or West may find it tough going.

[1] "It was George [Kaufman] who pointed out that you could always hold good cards merely by sitting South. 'No matter who writes the books or articles,' he said,

South holds the most terrific cards that I ever saw. There is a lucky fellow if I ever saw one.' Ever since then, I have always sat South.

That is the secret of my success, and I pass it along to you for whatever it is worth." Charles Goren

 

 

[hrothgar]

No

[AL78]

I have recorded my average HCP playing in my virtual club on BBO. I nearly always play North or East. There was initially a bias towards me picking up sub-average HCP over a session, and more so my partnership picking up sub-average HCP, but this has gone with having better than average HCP over the last four sessions. My conclusion so far is that what feels like a bias at the time is really a temporary streak, and the streaks can go both ways. Streaks are routine even in completely random systems.

[pilowsky]

Correct. Just thought I would put it to rest with actual evidence. Even though it should be obvious.

[awm]

While I don’t think there’s a bias in the deals, there does seem to be bias in the direction better players / more established pairs sit in the main bridge club.

 

Practicing with the same two pairs over several weeks, we have noticed that the pair sitting E/W normally “wins” because the E/W field is substantially weaker than the N/S field.

[paulg]

Some club members have told me that they prefer South because it looks more natural to have North at the top of the screen, West on the left and East on the right. This seems more true when they first start playing on BBO.

 

At my club in real-life, the stronger pairs sit North-South through preference since they control the boards and the Bridgemates and don't trust the weaker players to do so. It may mean that they are less likely to win (for the reasons outlined by Adam) but this seems less important!

 

This element of control for N/S probably transfers subconsciously from real life to BBO, even though it does not exist.

[hrothgar]

Correct. Just thought I would put it to rest with actual evidence. Even though it should be obvious.

 

But you haven't

 

You sample size is grossly insufficient and a bunch of the claims that you are making distract from your main point.

 

For example, if you want to show that HCPs are balanced, show the distribution of HCPs.

Don't use the number of tricks taken in double dummy contracts.

 

1. People can argue about whether double dummy solvers are accurate

2. People can argue about whether tricks taken is a good way to measure HCPs

 

Note: I don't disagree with your conclusions, but this isn't a very effective way to support them.

[pilowsky]

But you haven't

 

You sample size is grossly insufficient and a bunch of the claims that you are making distract from your main point.

 

For example, if you want to show that HCPs are balanced, show the distribution of HCPs.

Don't use the number of tricks taken in double dummy contracts.

 

1. People can argue about whether double-dummy solvers are accurate

2. People can argue about whether tricks taken is a good way to measure HCPs

 

Note: I don't disagree with your conclusions, but this isn't a very effective way to support them.

 

Dear Editor -

Thank you for your comments - I've been waiting for my whole career to able to write this response. You are an imbecile. You clearly know nothing about sampling or statistics. I have used at least double the sample size needed as per the Nyquist theorem.

Do you have the vaguest idea what you are talking about, or do you just make this stuff up as you go along? Just asking.

1. The robots were given random hands to play and random results were collected. They are not double-dummy. There is only one dummy in this conversation.

2. Argue till you you are blue in the face. In fact purple with rage. I don't care. The result is the important thing, not the tricks. It is the 'barometer' as Bridge players like to mangle the language.

 

To address the question of better players sitting EW. Further experiments were undertaken. This time deals from the Vugraph archive were randomly chosen by BBO (perhaps the computer wants to make a point?).

The results are the same. By the way, Bob Hamman a notoriously weak player sat South in one of the games.

[shyams]

ROFL. This is .... (I'm speechless).

 

What a moron!

[smerriman]

Suppose the deals were biased and South was dealt 13 spades every hand. N/S would get 0 IMPs every hand. So that proves the deals are generated at random, right?

[pilowsky]

Suppose the deals were biased and South was dealt 13 spades every hand. N/S would get 0 IMPs every hand. So that proves the deals are generated at random, right?

 

I paid for advanced robots.I set up a table. I put a robot on every seat. They all started playing with each other! After 12 boards I collected the data and started again.

It is all random.

What do Spades have to do with it?

 

 

[Vampyr]

Some club members have told me that they prefer South because it looks more natural to have North at the top of the screen, West on the left and East on the right. This seems more true when they first start playing on BBO.

 

At my club in real-life, the stronger pairs sit North-South through preference since they control the boards and the Bridgemates and don't trust the weaker players to do so. It may mean that they are less likely to win (for the reasons outlined by Adam) but this seems less important!

 

This element of control for N/S probably transfers subconsciously from real life to BBO, even though it does not exist.

 

How do the stronger pairs get the North-South seats? Is there some system besides first come first served?

[smerriman]

I paid for advanced robots.I set up a table. I put a robot on every seat. They all started playing with each other! After 12 boards I collected the data and started again.

It is all random.

What do Spades have to do with it?

The number of IMPs each side get is completely unrelated to whether dealing is biased or not. You stated that the end average being 0 means there is no bias, but that's 100% false - it would also be 0 if there *was* bias, like in my example.

[paulg]

How do the stronger pairs get the North-South seats? Is there some system besides first come first served?

It was first come, first served but traditionally certain places were left for particular pairs because they'd be upset if they had to sit elsewhere. It was, and is, an accommodating friendly club.

 

Now we toss for direction on most nights or draw for position in sim pairs.

[hrothgar]

Dear Editor -

Thank you for your comments - I've been waiting for my whole career to able to write this response. You are an imbecile. You clearly know nothing about sampling or statistics. I have used at least double the sample size needed as per the Nyquist theorem.

Do you have the vaguest idea what you are talking about, or do you just make this stuff up as you go along? Just asking.

 

Listen shiite for brains, I know that you're a short timer, don't know anyone on the forums, and seems to have some need to go and swing your dick around, so here's a bit of background information

 

I have multiple graduate degrees in this stuff including two from MIT.

(I graduated from there with almost a 5.0 average)

 

The job that I held before this one was the product manager for MATLAB's statistics system.

The job that I currently hold is Principle Data Scientist at Akamai

 

As for your "contributions" to this thread.

 

The Nyquist–Shannon sampling theorem comes out of signal processing. It describes the relationship between the frequency of a signal and the sample rate. It doesn't get used to determine the sample size for observational studies. It doesn't get used for classical power calculations.

 

> The robots were given random hands to play and random

> results were collected. They are not double-dummy.

> There is only one dummy in this conversation.

 

I agree.

It is the person who doesn't know that GIB uses double dummy solvers to determine its line of play

[pilowsky]

The number of IMPs each side get is completely unrelated to whether dealing is biased or not. You stated that the end average being 0 means there is no bias, but that's 100% false - it would also be 0 if there *was* bias, like in my example.

 

Incorrect. The number of IMPS is the only important thing. The scoreline is what matters; in the end, nothing else matters - that is what points schmoints means. That is why it's an interesting game.

 

Also, your use of the adverb 'completely' is ludicrously and ridiculously unscientific and will not be permitted. Two pairs of robots each alike in dignity play the same 6 sets of 12 boards.

In the end, NS achieves the same aggregate result as EW.

Therefore, there is no bias in the way that the hands are dealt. quod erat demonstrandum.

As to the question of whether or not stronger players prefer to sit EW, If this were true then there would be a bias in the outcome compared with the results as compared with that achieved by top players in Vugraph competitions.

There was not so there isn't.

It is basically the same experiment that Francis Galton did to prove that prayer does not work. The royal family does not live longer on average than other people, yet people are praying for them all the time. "God save the King" etc. Of course, I never meant it when I said it. I was probably not alone.

 

 

 

[hrothgar]

Note what happened when you repeated your little experiment

(I had to tweak the repeated 12.5 value because a ks.test doesn't like ties in a continuous distribution)

At a 10% significance level, a lot of your distributions are looking mighty different from one another

 

(Recall my earlier point about your sample sizes being grossly insufficient?)

 

> foo <- c(21.5, 24.5, 18.5, 32.5, 4.5, 10)

> bar <- c(12.5, 8, 18, 43.5, 31, 55)

> foo1 <- c(20.3, 23.3, 12.5001, 9.3, 1.2, 19.1)

> bar1 <- c(14.8, 6.8, 21.8, 32.4, 12.50001, 4.8)

 

> ks.test(foo, bar)

 

 

Two-sample Kolmogorov-Smirnov test

 

data: foo and bar

D = 0.33333, p-value = 0.9307

alternative hypothesis: two-sided

 

> ks.test(foo, bar1)

 

Two-sample Kolmogorov-Smirnov test

 

data: foo and bar1

D = 0.33333, p-value = 0.9307

alternative hypothesis: two-sided

 

> ks.test(foo1, bar)

 

Two-sample Kolmogorov-Smirnov test

 

data: foo1 and bar

D = 0.5, p-value = 0.474

alternative hypothesis: two-sided

 

> ks.test(foo1, bar1)

 

Two-sample Kolmogorov-Smirnov test

 

data: foo1 and bar1

D = 0.16667, p-value = 1

alternative hypothesis: two-sided

 

> ks.test(bar,bar1)

 

Two-sample Kolmogorov-Smirnov test

 

data: bar and bar1

D = 0.33333, p-value = 0.9307

alternative hypothesis: two-sided

 

> ks.test(foo,foo1)

 

Two-sample Kolmogorov-Smirnov test

 

data: foo and foo1

D = 0.33333, p-value = 0.9307

alternative hypothesis: two-sided

[hrothgar]

 

As to the question of whether or not stronger players prefer to sit EW, If this were true then there would be a bias in the outcome compared with the results as compared with that achieved by top players in Vugraph competitions.

There was not so there isn't.

 

 

foo <- c(21.5, 24.5, 18.5, 32.5, 4.5, 10)

bar <- c(12.5, 8, 18, 43.5, 31, 55)

 

mean(foo)

sd(foo)

mean(bar)

sd(bar)

ks.test(foo, bar)

 

> mean(foo)

[1] 18.58333

> sd(foo)

[1] 10.09166

> mean(bar)

[1] 28

> sd(bar)

[1] 18.53915

 

Two-sample Kolmogorov-Smirnov test

 

data: foo and bar

D = 0.33333, p-value = 0.9307

alternative hypothesis: two-sided

[hrothgar]

How do the stronger pairs get the North-South seats? Is there some system besides first come first served?

 

I suspect that the proposed dynamic is that

 

Players who serve tables are more likely to be established pairs AND

Players who serve tables are more likely to sit N/S

[hrothgar]

The number of IMPs each side get is completely unrelated to whether dealing is biased or not. You stated that the end average being 0 means there is no bias, but that's 100% false - it would also be 0 if there *was* bias, like in my example.

 

He's also using a peculiar amount of abstraction

 

Initially he is averaging together board results into six sets of 12 boards

And then he is calculating whether wins more IMPs and codes this as a -1 or a +1

And then he is summing across the number of boards

 

One wonders what would have happened had he chosen and odd number of hands.

[Vampyr]

It was first come, first served but traditionally certain places were left for particular pairs because they'd be upset if they had to sit elsewhere. It was, and is, an accommodating friendly club.

 

Now we toss for direction on most nights or draw for position in sim pairs.

 

It doesn’t sound that friendly or accommodating if there are people who throw a hissy fit if they don’t get to sit where you want.

 

I suspect that the proposed dynamic is that

 

Players who serve tables are more likely to be established pairs AND

Players who serve tables are more likely to sit N/S

 

In real bridge the management and/or director serves all the tables,

[hrothgar]

In real bridge the management and/or director serves all the tables,

 

Online bridge is not the same as F2F bridge.

 

At the most basic level, the overwhelming majority of games don't have "management", nor are there directors.

[smerriman]

Incorrect. The number of IMPS is the only important thing. The scoreline is what matters; in the end, nothing else matters - that is what points schmoints means. That is why it's an interesting game.

 

Also, your use of the adverb 'completely' is ludicrously and ridiculously unscientific and will not be permitted. Two pairs of robots each alike in dignity play the same 6 sets of 12 boards.

In the end, NS achieves the same aggregate result as EW.

Therefore, there is no bias in the way that the hands are dealt. quod erat demonstrandum.

OK, you cannot seriously be anything other than trolling this forum now if you can't even follow basic logic. But I'll try once more, who knows why. Here are some very basic questions for you to answer.

 

If IMPs averaging to 0 proves the lack of bias, who do you think will get more IMPs if there is a bias? The side that gets dealt more HCP? What if the HCP are distributed evenly, but the hands are always Goulash-style? Why would Goulash not average to 0?

 

In particular, who do you think will get more IMPs if there is a bug in the algorithm that will result in South getting dealt 13 spads every hand?

 

N/S will make 7S. Every other N/S will make 7S. N/S will receive exactly 0 IMPs.

 

No matter *what* the dealing algorithm is, the average number of IMPs N/S will receive per hand is 0, because IMPs solely measure how N/S compare to all other N/S at other tables, no matter how bad/good the hand. That's the whole point of duplicate scoring.

 

Or take daylongs. There is clearly a bias there, since South is always given the best hand. Are you saying the human scores in a daylong should average more than 0 IMPs? Because last I checked, that's mathematically impossible.

[pilowsky]

Listen shiite for brains, I know that you're a short timer, don't know anyone on the forums, and seems to have some need to go and swing your dick around, so here's a bit of background information

 

I have multiple graduate degrees in this stuff including two from MIT.

(I graduated from there with almost a 5.0 average)

 

The job that I held before this one was the product manager for MATLAB's statistics system.

The job that I currently hold is Principle Data Scientist at Akamai

 

As for your "contributions" to this thread.

 

The Nyquist–Shannon sampling theorem comes out of signal processing. It describes the relationship between the frequency of a signal and the sample rate. It doesn't get used to determine the sample size for observational studies. It doesn't get used for classical power calculations.

 

> The robots were given random hands to play and random

> results were collected. They are not double-dummy.

> There is only one dummy in this conversation.

 

I agree.

It is the person who doesn't know that GIB uses double dummy solvers to determine its line of play

 

I'm really delighted to discover that Akamai has decided to employ a data scientist with principles.

 

KS is not the best test. Better is the G-test for independence. I simplified it for the Forum so that a comparison could be made with Chi-squared. Neither are necessary because the result is bleeding obvious.

It does not matter what underlying software process GIB uses. What GIB does do is simulate.

You have been told this multiple times on the forum. Please listen. I only had to be told once but I am a fast learner. It may use the DD to simulate, but that does not mean that the 'simulation' is correct.

You are only 'correct' until your opponent makes their next move. This is true for all equilibrium games. It is true in life as well. Everything is 'obvious' when you know what happened.

 

Recently several people on the Forum complained that the deals were biased on BBO.

Think about what they mean. Did they mean that they did not they they did not get enough Aces? enough Kings enough spades? Wake up. What did they mean?

To help people you need to step into their shoes to answer their question. What people are saying is: When I play against other people of roughly equal ability and I am sitting EWN or South I seem to lose more often.

These people are NOT asking a mathematical question they are asking something else entirely. They are getting bad scores and they are blaming it on the cards that they are being dealt or the seat they are sitting in or the weather or the Jews or something else equally nonsensical. This is the reason Trump got elected. Because people thought they could solve all their problems by magic.

 

 

What I have done here is to demonstrate empirically - that there is no ghost in the machine. Pit four robots against each other and it does not matter where they are seated EW is just as likely to come out on top as NS.

 

 

Being competent in one area does not ensure competence in every area.

[pilowsky]

OK, you cannot seriously be anything other than trolling this forum now if you can't even follow basic logic. But I'll try once more, who knows why. Here are some very basic questions for you to answer.

 

If IMPs averaging to 0 proves the lack of bias, who do you think will get more IMPs if there is a bias? The side that gets dealt more HCP? What if the HCP are distributed evenly, but the hands are always Goulash-style? Why would Goulash not average to 0?

 

In particular, who do you think will get more IMPs if there is a bug in the algorithm that will result in South getting dealt 13 spads every hand?

 

N/S will make 7S. Every other N/S will make 7S. N/S will receive exactly 0 IMPs.

 

No matter *what* the dealing algorithm is, the average number of IMPs N/S will receive per hand is 0, because IMPs solely measure how N/S compare to all other N/S at other tables, no matter how bad/good the hand. That's the whole point of duplicate scoring.

 

Or take daylongs. There is clearly a bias there, since South is always given the best hand. Are you saying the human scores in a daylong should average more than 0 IMPs? Because last I checked, that's mathematically impossible.

 

Stephen, you throw the term 'troll' around but clearly do not know what it means. A 'troll' is a person whose sole motivation is to inflame and cause damage. What I am doing is stimulate conversation, education and assist people. Unlike many on this forum I never 'characterise', I am never rude, dismissive or abusive. When I provide explanations I do so within the limits of my knowledge base. I do not stroll outside of them.

When I want to learn something I ask questions.

Sometimes I ask questions that appear 'incompetent'. If I knew the answer, why would I ask the question? Only people that are trying to show off pose problems or ask questions when they already know the answer.

 

There is confusion about the usage of the word troll. The true troll is like an arsonist the have a psychopathology. They just want to see pain and suffering - that is how they get pleasure.

This is being misused in the media to refer to 'Russian troll farms'. These are not trolls. these are disinformation units of the FSB (formerly KGB) which reports to Putin (formerly head of the KGB).

Their actions are very purposeful. They want to destroy the government in the USA.

 

Obviously the strength in Daylongs is biased. What my 'little' experiment proves is that the hands dealt in the rest of BBO world are not hands re-used from daylongs. If they were then South would definitely get higher HCP counts.

This 'little' experiment should put peoples minds at ease.