[debrose]

 JLOGIC, on 2013-February-08, 11:01, said:

Bob also has a specific super accept for 3433 with 4 of partners suit, and we play over that that 3N is natural from partner (whereas 3n over other super accepts is not natural).

The system that han was talking about that I learned from bob was to solve the problem that 2N 3H 3S 3N must be corrected with 3 trumps since partner might be unbalanced (including even 5-5). This is bad when partner is 5332 and opener is 4333. A solution (if you don't play puppet) is to play 2N 3C 3D 3H (smolen) as not promising 4 hearts. Now opener with 3 spades bids 3S and responder can just bid 4 spades with 5 spades unbalanced. This leaves 2N 3H 3S 3N as a true choice when opener has 3 spades.

This might seem like overkill and placing too much importance on this issue, but that is how important it is imo.

Justin,
The solution you describe above didn't address how you handle the unbalanced 5-card spade hand when partner bids 3H instead of 3D. Presumably, you must bid 3S (either 5 spades unbalanced or artificial heart slam try?), so this system does have the drawback of wrong-siding spades sometimes.

[JLOGIC]

 debrose, on 2013-February-08, 22:20, said:

Justin,
The solution you describe above didn't address how you handle the unbalanced 5-card spade hand when partner bids 3H instead of 3D. Presumably, you must bid 3S (either 5 spades unbalanced or artificial heart slam try?), so this system does have the drawback of wrong-siding spades sometimes.

Yup definitely true.

[han]

 rhm, on 2013-February-08, 08:26, said:

Out of curiosity I just ran a simulation giving North 5♠332 and South 4♠333 and exactly 25 HCP together.
3NT made on 639 deals while 4♠ made double dummy on 471 deals.

For what it is worth bluecalm helped me analyse vugraph hands from elite pairs (as he likes to call them) with these two shapes and one table in 4M and the other in 3NT. Although the number of hands was not very large, 3NT was a big winner.

[fromageGB]

While I can intuitively see that with a marginal game 3NT is better than 4♠ because NT makes on average 8.7 tricks and spades 9.4, I am not convinced when there are a few more points in the hands. For example, with a 27 count, perhaps the probabilities change to NT=9.3 and ♠=10.1. Now you get 3NT making 9 for 400 and 4♠ making 10 for 430.

At IMPs it would be better for the security of the NT contract, but at matchpoints 4♠ is the winner, even if 1 in 3 of the NT contracts makes 10 tricks. I think this means that in MP pairs you should be concerned about this issue only when game is borderline.

[Fluffy]

 han, on 2013-February-09, 03:34, said:

For what it is worth bluecalm helped me analyse vugraph hands from elite pairs (as he likes to call them) with these two shapes and one table in 4M and the other in 3NT. Although the number of hands was not very large, 3NT was a big winner.

So we have clarified the 5332, next step is the 5422 I guess

[gnasher]

 han, on 2013-February-09, 03:34, said:

For what it is worth bluecalm helped me analyse vugraph hands from elite pairs (as he likes to call them) with these two shapes and one table in 4M and the other in 3NT. Although the number of hands was not very large, 3NT was a big winner.

That's certainly worth a lot more than a double-dummy simulation.

[rhm]

 gnasher, on 2013-February-09, 18:19, said:

That's certainly worth a lot more than a double-dummy simulation.

Why?
I said that double dummy simulation favored 4M and I stand by it.
So if double dummy simulation shows a profit for 3NT a few deals from real life would not change perception.

Rainer Herrmann

[rhm]

 fromageGB, on 2013-February-09, 05:26, said:

While I can intuitively see that with a marginal game 3NT is better than 4♠ because NT makes on average 8.7 tricks and spades 9.4, I am not convinced when there are a few more points in the hands. For example, with a 27 count, perhaps the probabilities change to NT=9.3 and ♠=10.1. Now you get 3NT making 9 for 400 and 4♠ making 10 for 430.

At IMPs it would be better for the security of the NT contract, but at matchpoints 4♠ is the winner, even if 1 in 3 of the NT contracts makes 10 tricks. I think this means that in MP pairs you should be concerned about this issue only when game is borderline.

I believe you are wrong.
Summary double dummy simulations are not as easy to interpret for matchpoints than for IMPs.
For the sake of the argument if you make say 0.2 tricks more on average over a large number of deals in a major suit game contract than in 3NT it is better playing notrump because in the majority of deals you will make the same number of tricks in both contracts.
Obviously you need to make more than 0.5 tricks per deal on average before it is worthwhile playing a major suit game instead of 3NT.
But even that is too low, because single dummy declarer makes about 0.2 tricks more in 3NT than double dummy, while there is no such difference between single dummy and double dummy in a major suit game.
This difference is probably due to the fact that the opening lead in 3NT is on average more crucial than in a trump contracts and no defense finds the right lead all the time.
So my yardstick that 4M should be preferred over 3NT in double dummy simulation is, if the average trick difference is at least 0.7.
This of course is a simplification, since on some deals there is more than a one trick difference between 3NT and 4M, but I do not think this has a big impact on the numbers and again it is more likely to favor 3NT. If you are making less tricks in 3NT at matchpoints it is of little consequence for your matchpoint score whether you are off by one or more tricks.

Now it is well known that when the total combined HCP holding rises the chances getting an additional tricks from trumps tends to diminish, again favoring 3NT.
I repeated the simulation but with 27 HCP combined.

The result over 1000 deals:

3NT made on 879 deals while 4♠ made on 846 deals
Average number of tricks in 3NT was 9.628 while the number of trick in ♠ 10.145

As expected the trick difference dropped from 0.7 in my previous simulation with 25 HCP combined to 0.5
According to my yardstick this does not justify to prefer the major in the long term matchpoint wise.


Rainer Herrmann

[gnasher]

 rhm, on 2013-February-10, 03:12, said:

Why?
I said that double dummy simulation favored 4M and I stand by it.
So if double dummy simulation shows a profit for 3NT a few deals from real life would not change perception.

Because a double-dummy simulation represents double-dummy play, and real-life results reflect real-life play. When designing bidding methods, I'm interested in the results that I would get in real life, not the results that I would get if everyone could see all the cards.

Here are some reasons why double-dummy analysis might, in this situation, provide misleading results:
- In four of the major we may have time to try several different finesses, whereas in 3NT we may have to guess which finesse to take. DD analysis will assume that we would get the guess right.
- In four of the major we may be able to avoid a guess by an elimination. DD analysis assumes that we would get the guess right in 3NT.
- In 3NT, with the declaring side 3-3 or 3-2 in all the side-suits, it will quite often be right to make a short-suit lead. DD analysis assumes that the defenders would do this. Whether they would do at the table depends on the auction: on an unrevealing auction like 1NT-2♥;2♠-3NT it will be hard for the opening leader to do this; on an auction where the declaring side announce a fit, discover that opener is 4333, and then settle in 3NT, it will be easier.
- Against four of the major real-life defenders may choose to make an attacking lead, whereas double-dummy defenders know that there are no discards to be had, so no need to attack.

I don't know how much these and other factors affect the relationship between double-dummy and single-dummy results, but I'm certain that it's wrong to extrapolate from double-dummy to single-dummy without considering them.

[fromageGB]

 rhm, on 2013-February-10, 04:04, said:

I repeated the simulation but with 27 HCP combined.
The result over 1000 deals:
3NT made on 879 deals while 4♠ made on 846 deals
Average number of tricks in 3NT was 9.628 while the number of trick in ♠ 10.145

As expected the trick difference dropped from 0.7 in my previous simulation with 25 HCP combined to 0.5
According to my yardstick this does not justify to prefer the major in the long term matchpoint wise.

Thanks for the new simulation result, this does help. Applying matchpoints to your figures, assuming when you go off you go one off, and when making you make 9 or 10 in NT or 10 or 11 in spades, I get the expected not vulnerable matchpoint score of 413 for 3NT and 417 for 4♠. Interesting. If vulnerable, expected 607 for 3NT and 609 for 4♠.

Almost identical, and as there are likely to be one or two contracts down 2 in spades, and a few plus 2 in NT, it pushes the preferred contract to be 3NT. As you say, as total hcp goes higher than 27, it will favour NT even more.

So you have persuaded me, 3NT is always to be preferred.

[rhm]

 gnasher, on 2013-February-10, 05:06, said:

Because a double-dummy simulation represents double-dummy play, and real-life results reflect real-life play. When designing bidding methods, I'm interested in the results that I would get in real life, not the results that I would get if everyone could see all the cards.

Here are some reasons why double-dummy analysis might, in this situation, provide misleading results:
- In four of the major we may have time to try several different finesses, whereas in 3NT we may have to guess which finesse to take. DD analysis will assume that we would get the guess right.
- In four of the major we may be able to avoid a guess by an elimination. DD analysis assumes that we would get the guess right in 3NT.
- In 3NT, with the declaring side 3-3 or 3-2 in all the side-suits, it will quite often be right to make a short-suit lead. DD analysis assumes that the defenders would do this. Whether they would do at the table depends on the auction: on an unrevealing auction like 1NT-2♥;2♠-3NT it will be hard for the opening leader to do this; on an auction where the declaring side announce a fit, discover that opener is 4333, and then settle in 3NT, it will be easier.
- Against four of the major real-life defenders may choose to make an attacking lead, whereas double-dummy defenders know that there are no discards to be had, so no need to attack.

I don't know how much these and other factors affect the relationship between double-dummy and single-dummy results, but I'm certain that it's wrong to extrapolate from double-dummy to single-dummy without considering them.

This has been raised often, but double dummy analysis is apparently still poorly understood.
Yes double dummy analysis is not real Bridge, because real Bridge is about single dummy analysis and human errors.

However, what has been known for a long time is that over a number of deals the result of double dummy analysis is close to single dummy analysis.
Double dummy results are a very good proxy for single dummy results.
This happens because the differences you mention cancel each other out. Neither defense nor declarer play is double dummy.

What difference remains can be quantified.
Low notrump contracts slightly favor declarer single dummy, while slam contracts slightly favor the defense.
For 3NT declarer's advantage is 0.2 tricks, for 4M it is zero.
Extrapolating from a small sample of real life plays is problematic, because

a) a small sample may not be representative.
b) you have little control what extraneous information there was available at the time of the decision at the table and you will not get these specific deals in your lifetime.

Rainer Herrmann

[gnasher]

 rhm, on 2013-February-10, 07:33, said:

However, what has been known for a long time is that over a number of deals the result of double dummy analysis is close to single dummy analysis.
Double dummy results are a very good proxy for single dummy results.
This happens because the differences you mention cancel each other out. Neither defense nor declarer play is double dummy.

What difference remains can be quantified.
Low notrump contracts slightly favor declarer single dummy, while slam contracts slightly favor the defense.
For 3NT declarer's advantage is 0.2 tricks, for 4M it is zero.

I have given you reasons why double-dummy results may not accurately model single-dummy results in this particular situation, that is specifically when we hold a 5=3=3=2 opposite a 4=3=3=3. Do you have evidence that for this specific situation these factors cancel out? If so, I think you should share this evidence.

[gnasher]

 rhm, on 2013-February-10, 07:33, said:

This has been raised often, but double dummy analysis is apparently still poorly understood.

You seem to be confusing scepticism with a lack of understanding.

I'm not sure what leads you to believe that I don't understand double-dummy analysis. I understand it very well. In this instance I simply don't accept your conclusions, given the evidence that you have so far provided.

This post has been edited by gnasher: 2013-February-10, 08:10

[rhm]

 gnasher, on 2013-February-10, 07:50, said:

I have given you reasons why double-dummy results may not accurately model single-dummy results in this particular situation, that is specifically when we hold a 5=3=3=2 opposite a 4=3=3=3. Do you have evidence that for this specific situation these factors cancel out? If so, I think you should share this evidence.

The reason is if double dummy works as a good proxy for the expected result single dummy in general, there is no Bridge (or statistical) reason why the results should somehow be different when making some specific Bridge assumptions about distribution and HCP, but which are otherwise not related to the play of the hand itself.
What matters is only whether the number of deals in the simulation is high enough to give an approximation about the actual result for the total number of possible deals where this scenario can exist.

 gnasher, on 2013-February-10, 07:55, said:

You seem to be confusing scepticism with a lack of understanding.

I'm not sure what leads you to believe that I don't understand double-dummy analysis. I understand it very well. In this instance I simply don't accept your conclusions, given the evidence that you have so far provided.

I am sure you know what double dummy analysis is bridge wise. The argument you use is Bridge related when the main argument is related to statistics, not Bridge.
You claim double dummy analysis is not real Bridge. Therefore you have little confidence in double dummy simulations.
But nobody claims this to be the case.
The claim is only that the result of double dummy analysis over many boards approaches single dummy real life results, assuming players have a desire to do well and have some competency for the game.
Therefor we can use double dummy analysis to tell us what will happen single dummy over many boards.
It does not matter whether a double dummy analysis will give the same result as single dummy on any specific deal, nor whether the result is obtained by the same method.
That is not the case but also not the point.

Rainer Herrmann

[gnasher]

 rhm, on 2013-February-10, 08:53, said:

The reason is if double dummy works as a good proxy for the expected result single dummy in general, there is no Bridge (or statistical) reason why the results should somehow be different when making some specific Bridge assumptions about distribution and HCP, but which are otherwise not related to the play of the hand itself.
What matters is only whether the number of deals in the simulation is high enough to give an approximation about the actual result for the total number of possible deals where this scenario can exist.

Of course the constraints can affect how well the double-dummy simulation models real-life. Suppose that I carry out a simulation where I specify that the declaring side has Axxx opposite KJ109. Would you expect the benefit of being double-dummy at notrumps to be the same as the benefit of being double-dummy with this suit as trumps? Obviously, it isn't, because in a notrump contract you may be able to delay playing this suit until you know how everything else breaks, but in a suit contract you can't. Hence in this example the advantage of being double-dummy in a suit would be greater than the advantage of being double-dummy in notrumps.

Quote

I am sure you know what double dummy analysis is bridge wise. The argument you use is Bridge related when the main argument is related to statistics, not Bridge.
You claim double dummy analysis is not real Bridge. Therefore you have little confidence in double dummy simulations.

No, I say that in this situation double-dummy analysis may not accurately reflect real bridge, because in this situation the ratio

(Advantage of being double-dummy at suit) : (Advantage of being double-dummy at notrumps)

May be different from the same ratio calculated over all hands, or calculated for a different set of hands.

Quote

But nobody claims this to be the case.
The claim is only that the result of double dummy analysis over many boards approaches single dummy real life results.
Therefor we can use double dummy analysis to tell us what will happen single dummy over many boards.
It does not matter whether a double dummy analysis will give the same result as single dummy on any specific deal.

Which part of that do you think that I didn't already understand?

[cherdano]

My judgement is that double-dummy in this situation favours the defenders at 3N, since the lead is so important: I think there will be many hands where declarer has 9 tricks of the top, and defenders 5, or where each side can set up that many tricks by losing the lead once.
So if double-dummy results favour playing 3N, then I think in real life the case is even stronger.

But my judgement already told me that playing 3N with 5332 opposite 4333 is better than playing 4M in your 5-4 fit, so I am not sure how much the double-dummy simulation helps.

Han's analysis, on the other hand, has the obvious problem of selection bias - it tells us that on hands where some world class pair chose to bid 3N, the 3N contract did better. That tells us that we would do well to imitate their strategies in choosing 3N, but unless what that strategy is, that's not all that helpful either. (Except for winning arguments on BBF against those who think you are wasting your time trying to find 3N on those hands.)

My views are that:
- If all you tell me are the shapes, and I can choose to play 3N or 4M in the 5-4 fit, on an auction that reveals nothing but the 5-card suit opposite a 1N opener, then choosing 3N is far superior.
- On some hands, the 4333 hand should choose to play 4M - e.g. with a xxx side suit, especially if he has already announced the fit.

I don't see how either the simulation or Han's analysis helps me much to validate or refute these views.

[GreenMan]

 cherdano, on 2013-February-10, 11:00, said:

Han's analysis, on the other hand, has the obvious problem of selection bias - it tells us that on hands where some world class pair chose to bid 3N, the 3N contract did better.

It also tells us that on hands where some other world class pair chose to bid 4M, the 4M contract did worse.

[phoenix214]

It is technicly possible to avoid such fits if you adjusst stayman a bit. 2D says "I dont have a 4 card major or I have any 4333". And with some 5332 you just bid stayman so you can play 3NT on bad fits

[han]

I agree with Arend that the vugraph results are biased and it is hard to draw conclusions from them. The double dummy results are perhaps more reliable, even though they are double dummy. Fortunately both say the same thing: with 4-3-3-3 vs 5-3-3-2 you should play 3NT, even if you have a 5-4 fit.

Another possible test is to force a computer program to play lots of hand with these conditions, both in 3NT and 4M. I don't have a computer program that can do that.

[han]

 phoenix214, on 2013-February-11, 00:57, said:

It is technicly possible to avoid such fits if you adjusst stayman a bit. 2D says "I dont have a 4 card major or I have any 4333". And with some 5332 you just bid stayman so you can play 3NT on bad fits

If you compare this route to the transfer followed by 3NT route (promising 5332 shape) then you give away more information and you do less well because you will miss some good 5-3 fits when opener is not 4333.