[raspeball]

Do anybody know if there exist analysis comparing how much the length of the auction affects the success rate of the final contract (Based on real life data with strong players)?

This could (easily?) be done by comparing tables where the same contract was reached, but where the length of the auction was different.

It makes sense that the shorter bidding sequences should score better on average, since less is revealed about the hands, making the defence harder.
(Of course, on the other side- Short and less revealing bidding, will cost in ending up in poorer contracts).

[hrothgar]

> This could (easily?) be done by comparing tables where the same contract
> was reached, but where the length of the auction was different.

Feels like you have a badly biased estimator

You're only looking at auctions where the two sides reach the identical contract. As such, the side that is basing enjoys all of the benefits but non of the costs (presumably less accurate auctions)

It might be better to look at a set of (non competitive) auctions where the hands were played in the same strain and try to use the length of the auction as a predictor of some kind...

[thepossum]

How about an even more basic estimator

- but there are some rather obvious confounds, and how many things would you need to control for in reality - maybe compare the number of bids in every slam contract or something like that - see if there is any effect at all - maybe controlling with similar tournament types and lengths too - but I suspect any effect would be hard to isolate

Maybe ask David atm - come to think of it where is he - drowned under all his scripts perhaps

Just a thought though about possible confounds, maybe also restrict it to a large database of experts playing top tournaments - not being one myself I am of the view that fewer bidding options and less complex systems are beneficial - with obvious confounds - you gave them far too much information with all those confused bids. I'm not sure you could isolate my hypothesised effect either - they could see you didn't have a clue but were probably better anyway. That is why I try to bluff my way with a direct and pre-emptive style

Will follow the discussion with interest

[DavidKok]

thepossum, on 2023-April-09, 18:59, said:

Maybe ask David atm - come to think of it where is he - drowned under all his scripts perhaps

If you were referring to me, I was waiting for people to comment on my code. I thought my lack of sharing the code was what kept people from contributing. Wasn't it?

I have no idea how to go about quantifying information leakage and accuracy. Some thoughts:

• Sharing more information wins by getting to better contracts, while blasting wins by making the lead and defence more difficult. On any given contract blasting has a clear edge - instead you'd need to compare the frequency where blasting leads to a suboptimal contract while more scientific bidding would have led to a superior alternative, to the frequency of slipups on defence.
  
• The old adage is 'bid game fast, bid slam slow'. I think accuracy in bidding will win at the slam level, while it won't be as relevant at the game level. Ideally you would split this analysis by the level of the contract (and good luck deciding what to do when one approach is in game while the other is in slam).
  
• Ideally you'd also split this analysis by skill of the players. Since one of the main relevant contributions is the rate of slip-ups of the defenders there are predictably differences when facing really strong defenders versus weaker defenders. Conversely, if we assume we are approximately as good as the rest of the field, if the field is weak that introduces a correlation - presumably our bidding agreements and hand evaluation are also weak, and our general bidding accuracy might be low regardless of the number of bids involved.
  
• Some attempts have been made in the past to quantify information density of different approaches - at the very least you could determine some frequency spreads of certain approaches, and try to invoke something like the useful space principle to help order them. As far as I know most of this has only been done informally though.

[thepossum]

DavidKok, on 2023-April-10, 03:15, said:

If you were referring to me, I was waiting for people to comment on my code. I thought my lack of sharing the code was what kept people from contributing. Wasn't it?

I have no idea how to go about quantifying information leakage and accuracy. Some thoughts:

• Sharing more information wins by getting to better contracts, while blasting wins by making the lead and defence more difficult. On any given contract blasting has a clear edge - instead you'd need to compare the frequency where blasting leads to a suboptimal contract while more scientific bidding would have led to a superior alternative, to the frequency of slipups on defence.
  
• The old adage is 'bid game fast, big slam slow'. I think accuracy in bidding will win at the slam level, while it won't be as relevant at the game level. Ideally you would split this analysis by the level of the contract (and good luck deciding what to do when one approach is in game while the other is in slam).
  
• Ideally you'd also split this analysis by skill of the players. Since one of the main relevant contributions is the rate of slip-ups of the defenders there are predictably differences when facing really strong defenders versus weaker defenders. Conversely, if we assume we are approximately as good as the rest of the field, if the field is weak that introduces a correlation - presumably our bidding agreements and hand evaluation are also weak, and our general bidding accuracy might be low regardless of the number of bids involved.
  
• Some attempts have been made in the past to quantify information density of different approaches - at the very least you could determine some frequency spreads of certain approaches, and try to invoke something like the useful space principle to help order them. As far as I know most of this has only been done informally though.

Hi David

I was just suggesting you were in full on analysis mode with access to a database of hands

I am interested in any serious discussion of Bridge statistics with anyone who has a clue

Some earlier attempts at raising the complexity of Bridge distributions and analysis have fallen on deaf ears

Hopefully what I was getting at covered some of what you and Richard are getting at

I haven't been motivated enough to get Double Dummy solver working with Mr Pavlicek's database

Is it possible to compare all 6 spade (oops a bit limiting -every 6M contract) contracts and auctions at a particular level of expertise

Thanks for the comment - encouraging to know I am not totally off in an irrelevant direction

Note some of my thoughts were likely influenced by Matthew too but etc

How do we drill down and work it out

But my view which I likely did not communicate well is that (from my statistical and Bridge knowledge) identifying such an effect could be difficult - even if broken down by skill and tournament type and other forms of control

To me having tried to get to grips over the last 6 years or so with the complexity of Bridge anaylsis its an interesting challenge - to even define the problem - crude estimator. Every 6M hand in Mr Pavilecek's data by skill level and tournament type and how it turned out by number of bids

EDIT I could be overcomplicating it. Take all the hands for the world teams and compare scores to number of bids - the IMPs/Bid metric - the only way I can possibly get a grip on the amount of control required - I imagine all the teams have their different coaching teams (offensive nd defensive) including an analytics team and a video analysis team - very secret too I imagine

Meeting with the senior coach - you need to reduce your number of bids by an average of 0.25 per hand and stop wiggling your little finger when nervous

[raspeball]

DavidKok, on 2023-April-10, 03:15, said:

If you were referring to me, I was waiting for people to comment on my code. I thought my lack of sharing the code was what kept people from contributing. Wasn't it?

I have no idea how to go about quantifying information leakage and accuracy. Some thoughts:

• Sharing more information wins by getting to better contracts, while blasting wins by making the lead and defence more difficult. On any given contract blasting has a clear edge - instead you'd need to compare the frequency where blasting leads to a suboptimal contract while more scientific bidding would have led to a superior alternative, to the frequency of slipups on defence.
  
• The old adage is 'bid game fast, bid slam slow'. I think accuracy in bidding will win at the slam level, while it won't be as relevant at the game level. Ideally you would split this analysis by the level of the contract (and good luck deciding what to do when one approach is in game while the other is in slam).
  
• Ideally you'd also split this analysis by skill of the players. Since one of the main relevant contributions is the rate of slip-ups of the defenders there are predictably differences when facing really strong defenders versus weaker defenders. Conversely, if we assume we are approximately as good as the rest of the field, if the field is weak that introduces a correlation - presumably our bidding agreements and hand evaluation are also weak, and our general bidding accuracy might be low regardless of the number of bids involved.
  
• Some attempts have been made in the past to quantify information density of different approaches - at the very least you could determine some frequency spreads of certain approaches, and try to invoke something like the useful space principle to help order them. As far as I know most of this has only been done informally though.

Thanks,
Another approach could be like this:

• Compare only hands from the knockout stages of major tournaments.
  
• Compare the result from a match.
  
• Use only hands with game going strength.(So each table reaches at least game)
  
• Use only hands where the opponents do not interfere to much (Maybe a simple overcall, or a double somewhere is allowed)
  
• See who wins the most imps, compared to how long the auction is.
  
• Try to analyze the data, and see if this indicates something.

Obviously there are lots of reasons why the result from such a study might be inaccurate, but it could be an indication of what strategies are successfull.

[DavidKok]

Personally I think it is difficult to answer this question, in part because it is too vague. The tradeoff of information leakage and accuracy is highly contingent on the auction thus far and the possible final contracts. If I wanted to try to get some insights into this question I would set up multiple proxy variables, and hope that sufficiently many point in the same direction on interesting deals. I'm not sure what these proxies would be - both the score of a deal and the number of bids taken are good examples, but I don't think they are sufficient.
As an aside, quite a few modern systems use more bids to convey less information. In Dutch Doubleton the sequence 1♣-1♦; 1♥-1♠; 1NT shows 12-13 balanced from opener without promising or denying any particular shape, while responder has shown something akin to a Walsh 1♦ response to 1♣. There are other system examples where a long route is taken to show a generic hand type, to free up direct bids for specific shapes.

It might be lower effort and more representative to focus on specific auctions, e.g. 'NS reach game or slam after they open a strong notrump'.

[mikeh]

I doubt that the question is susceptible to a meaningful answer, due to the various system philosophies out there

As an example, I have played and am now playing (in a more limited context) systems in which one partner, invariably for us, responder, implements a relay sequence

Back in the old days…late 1990s…we often had extremely long auctions….each call being alerted. I remember one where we alerted 13 times before placing the contract in a slam, declared by responder.

On being asked, at the end of the auction, I could say (and I no longer remember the exact hand) that he’s shown 3=1=5=4, with 5 controls. He has either both the AK of spades or neither of the,. Same in clubs. He has one of the A or K of diamonds. He has both the club and diamond queens but not the queen of spades. He has neither the Ace nor king of hearts, but might have the queen…I couldn’t ask.

He, in answering questions, could say: he has at least a decent 14 count, but other than that I know absolutely nothing about his hand other than that, opposite what I’ve told him, he thinks 6S is the correct contract.

Ok…lots of leakage there, in terms of opener’s hand, but the ambiguity in location of controls reduces that a bit, and of course having zero idea of my shape doesn’t help. On the opening lead having been made, the appearance of dummy can help with some reverse engineering but the reality is that our extremely long auction almost surely resulted in less information leakage than anything but the purest, and random, bashing would create.

So in relay, if the system is well designed such as to maximize the chances of the relayer being declarer, I don’t see much correlation between length of auction and information leakage.

Similarly, when Meckwell were the top pair in the world, they often had long auctions, in theory giving lots of information yet for many years they were arguably the best bidders, in terms of consistently reaching good contracts, in the world.

Finally (?), David’s point about skill level is pertinent.

I think that at the expert level the defenders are so skilled, and their methods so refined (I’m not including the ‘great’ defenders who used illegal methods) that there are few pure bashers. Maybe once in a while, but generally most top pairs like to bid slowly on, especially, slam and grand slam decisions. Otoh, many top pairs rarely, if ever, use game tries.

I think that if there’s one area of proper concern for cost-benefit analysis, it’s wrt to game tries. We, in my partnerships, do have various tries in our toolbox but rarely use them. The main exception is an auction such as 1D 1M 2M, where the raise might be on a weak hand with three trump…now the game try, asking min-max and trump length, seems worth the leakage.

Against lesser defenders, bashing is more apt to work, in absolute terms. But so too is slow bidding…the weaker the defenders, the less use they’ll make of information leakage anyway, and the bash can lead to silly contracts on occasion.

All in all, given the tremendous range of skills amongst players and partnerships, the huge variety of methods (especially at the expert level where long-standing partnerships will have many idiosyncratic treatments), I think there are too many variables to allow for significant findings to be made.

[raspeball]

mikeh, on 2023-April-10, 14:30, said:

I doubt that the question is susceptible to a meaningful answer, due to the various system philosophies out there

As an example, I have played and am now playing (in a more limited context) systems in which one partner, invariably for us, responder, implements a relay sequence

Back in the old days…late 1990s…we often had extremely long auctions….each call being alerted. I remember one where we alerted 13 times before placing the contract in a slam, declared by responder.

On being asked, at the end of the auction, I could say (and I no longer remember the exact hand) that he’s shown 3=1=5=4, with 5 controls. He has either both the AK of spades or neither of the,. Same in clubs. He has one of the A or K of diamonds. He has both the club and diamond queens but not the queen of spades. He has neither the Ace nor king of hearts, but might have the queen…I couldn’t ask.

He, in answering questions, could say: he has at least a decent 14 count, but other than that I know absolutely nothing about his hand other than that, opposite what I’ve told him, he thinks 6S is the correct contract.

Ok…lots of leakage there, in terms of opener’s hand, but the ambiguity in location of controls reduces that a bit, and of course having zero idea of my shape doesn’t help. On the opening lead having been made, the appearance of dummy can help with some reverse engineering but the reality is that our extremely long auction almost surely resulted in less information leakage than anything but the purest, and random, bashing would create.

So in relay, if the system is well designed such as to maximize the chances of the relayer being declarer, I don’t see much correlation between length of auction and information leakage.

Similarly, when Meckwell were the top pair in the world, they often had long auctions, in theory giving lots of information yet for many years they were arguably the best bidders, in terms of consistently reaching good contracts, in the world.

Finally (?), David’s point about skill level is pertinent.

I think that at the expert level the defenders are so skilled, and their methods so refined (I’m not including the ‘great’ defenders who used illegal methods) that there are few pure bashers. Maybe once in a while, but generally most top pairs like to bid slowly on, especially, slam and grand slam decisions. Otoh, many top pairs rarely, if ever, use game tries.

I think that if there’s one area of proper concern for cost-benefit analysis, it’s wrt to game tries. We, in my partnerships, do have various tries in our toolbox but rarely use them. The main exception is an auction such as 1D 1M 2M, where the raise might be on a weak hand with three trump…now the game try, asking min-max and trump length, seems worth the leakage.

Against lesser defenders, bashing is more apt to work, in absolute terms. But so too is slow bidding…the weaker the defenders, the less use they’ll make of information leakage anyway, and the bash can lead to silly contracts on occasion.

All in all, given the tremendous range of skills amongst players and partnerships, the huge variety of methods (especially at the expert level where long-standing partnerships will have many idiosyncratic treatments), I think there are too many variables to allow for significant findings to be made.

So maybe to get a meaningful result, you would need to know some details about what the bidding actually means?

A much simpler situation might be easier to analyze:

• Say you compare a hand where both sides opened 1nt(Could be both to the same contract, or to different contracts).
  
• Both sides bid to either 3nt or 4♥/♠.
  
• The two tables had different auctions.
  
• Possibly also add some restrictions to the distribution and combined total strength of the hands.
  
• Limit the amount of interference allowed (Only lead directing doubles)
  
• Allow the program do some interpretation on what the bidding actually means (So that it can recognize transfer auction, stayman and puppet stayman auctions)

[pescetom]

DavidKok, on 2023-April-10, 03:15, said:

If you were referring to me, I was waiting for people to comment on my code. I thought my lack of sharing the code was what kept people from contributing. Wasn't it?

Just to say belatedly thanks, I downloaded your scripts and read many of them, looked mostly ok.
I am still perversely optimistic/obstinate that BBO will restore DD calculation in Dealer, so I cannot run them or try alternatives at this moment.

[johnu]

mikeh, on 2023-April-10, 14:30, said:

Otoh, many top pairs rarely, if ever, use game tries.

IIRC it was Kit Woolsey who said his game try method was to bid game, and try to make it.

[the hog]

johnu, on 2023-April-13, 15:52, said:

IIRC it was Kit Woolsey who said his game try method was to bid game, and try to make it.

Look at Mats Nisland's posts on Bridge winners

[sfi]

johnu, on 2023-April-13, 15:52, said:

IIRC it was Kit Woolsey who said his game try method was to bid game, and try to make it.

Many people use that line - I don't think you're going to have much luck tracing its genesis.

[helene_t]

What about:
- deal 1000 deals consistent with 1n (p) 2n
- for each of them deal say 50 additional deals where the hand on lead is the same, and dd solve them
- for each of the original 1000 deals, choose the lead that is
- A: favoured by the entire set of 50 deals
- B: favoured by a set of 50 deals that provide the same information if dummy had to invite via stayman

OK there is a flaw namely that the stayman auction may not leak the info that dummy doesn't have a 4card major, but the correction of that can be left as an exercise to the reader 😉