Does transfer to the stronger hand gain tricks?

[pescetom]

I hacked the rough script below to see if a simple major suit transfer was gaining/costing tricks.

It assigns North an orthodox 15-17 1NT and assumes that one of NS will declare and play in spades any time South has 5+ spades.

#
# Does transfer to the stronger hand gain tricks?
#
produce 10000
N1NT = shape(north, any 4333 + any 4423 + any 5332) and hcp(north)>=15 and hcp(north)<=17
S5S = spades(south)>=5
Ntricks = tricks(north,spades)
Stricks = tricks(south,spades)
delta = Ntricks - Stricks
action frequency "of spades tricks" (Ntricks,8,13), frequency "of extra spades tricks due to transfer" (delta,-2,2)
condition N1NT and S5S

 

Here is one run:

Frequency of spades tricks:
Low	    1130    
   8	    1391
   9	    2054
  10	    2369
  11	    1854
  12	     932
  13	     270
Frequency of extra spades tricks due to transfer:
Low	       1
  -2	       4
  -1	     260
   0	    9159
   1	     559
   2	      16
High	       1
Generated 1190045 hands
Produced 10000 hands
Initial random seed 1618327595
Time needed  589.581 sec

 

From this run it looks like transfer is gaining a trick (occasionally two) 5.9% of times and costing a trick 2.6%.

[helene_t]

Pavlicek found that letting the stronger hand declare generally costs a slight fraction of a trick if the strength difference is moderate, while gaining a slight fraction of a trick when the strength difference is extreme. Some of the differences were statistically significant as he made a huge number of sims, but the differences were tiny.

 

Here, having the balanced hand declare may offer an additional advantage since if dummy has a singleton in the suit being led, declarer may have an honour to protect. On the other hand, with Hx opposite Hxx it may be better to have the 3-card suit in dummy. So I would expect it not to make any difference in your sims. Maybe Pavlicek did something wrong, or maybe your results are actually consistent with Pavlicek's results, as some of your hands will have a very weak dummy, and it may be important that declarer is balanced and dummy has a 5+ suit.

 

Single dummy, I would expect the effect to be larger as defenders will sometimes lead into a tenance unnecessarily, and sometimes they will lead a trump when the layout is something like

xx-KT9xxx-Qxx-Ax, of course the opposite could happen in a side suit but then it may be less likely to cost a trick as the side suit queen can sometimes be ruffed out.

 

On the other hand, playing transfers declarer will usually have a range of 15-17 when dummy may have 0-7 or 10-15, so it is easier for the defenders to place the honours when the more narrowly defined hand is declarer. The same sometimes applies to distributions although that obviously depends on methods. Smolen, for example, discloses less about the nt opener's shape than about partner's shape.

 

It would be nice to see some single dummy analysis, as this is a very important issue for system design: under what circumstances should system designers worry about right-siding? But this is obviously quite difficult because we would have to compare some specific systems (say standard transfers vs 2-way stayman) to be able to make assumptions about what information has been leaked.

[steve2005]

using double dummy spoils the statistics

often you blow a trick on the lead

whereas in double dummy you always get it right (like finding a suit for partner to lead thru declarer

[smerriman]

Yes, the whole argument behind rightsiding is that a) the defense are much more likely to give up a trick leading into the strong hand vs the weak, and b) it's easier to defend from trick 2 onwards when you can see more of declarer's values.

 

Neither of these are taken into account by double dummy analysis.

 

The only time you'll get a difference double dummy is when leading into all 4 suits gives up a trick. So when passive defense won't suffice, active defense requires immediately leading through declarer, and partner doesn't have an entry allowing you to do so (or that entry is crucial later). To be honest, I would have expected the results to be closer.

[LBengtsson]

a hand with more honour cards has more chance of performing a deceptive maneuver during the play than a hand with less, that is how I see this. so leading up to the strong hand can cost a trick for the defense, and not reading declarers hand correctly can probably also cost a trick also.

 

hiding shape is important for strong hand also. if the weaker hand does not transfer and becomes declarer, the defensers already know that the weaker hand has 5+ cards in the transfer suit, so working out hand shape is easier for defense. if the stronger hand becomes declarer, the defensers know nothing about level of trump fit, or declarers hand shape, except on a super accept (4 trumps and a max hand)

[pescetom]

a hand with more honour cards has more chance of performing a deceptive maneuver during the play than a hand with less, that is how I see this. so leading up to the strong hand can cost a trick for the defense, and not reading declarers hand correctly can probably also cost a trick also.

 

hiding shape is important for strong hand also. if the weaker hand does not transfer and becomes declarer, the defensers already know that the weaker hand has 5+ cards in the transfer suit, so working out hand shape is easier for defense. if the stronger hand becomes declarer, the defensers know nothing about level of trump fit, or declarers hand shape, except on a super accept (4 trumps and a max hand)

I had an amusing demonstration of this last night, when our (disclosed) agreements allowed my 16hcp 1354 to open 1nt and become declarer in 4♥ on a 3-5 fit. All LHO knew about my distribution was 3 card hearts and apparently he didn't put much stock in the announced possible singleton, as when on trick two I led low towards ♠K842 in dummy the K held. I then switched to clubs, won by RHO. His face when I ruffed his ♠J return was priceless.

[Winstonm]

Pavlicek found that letting the stronger hand declare generally costs a slight fraction of a trick if the strength difference is moderate, while gaining a slight fraction of a trick when the strength difference is extreme. Some of the differences were statistically significant as he made a huge number of sims, but the differences were tiny.

 

Here, having the balanced hand declare may offer an additional advantage since if dummy has a singleton in the suit being led, declarer may have an honour to protect. On the other hand, with Hx opposite Hxx it may be better to have the 3-card suit in dummy. So I would expect it not to make any difference in your sims. Maybe Pavlicek did something wrong, or maybe your results are actually consistent with Pavlicek's results, as some of your hands will have a very weak dummy, and it may be important that declarer is balanced and dummy has a 5+ suit.

 

Single dummy, I would expect the effect to be larger as defenders will sometimes lead into a tenance unnecessarily, and sometimes they will lead a trump when the layout is something like

xx-KT9xxx-Qxx-Ax, of course the opposite could happen in a side suit but then it may be less likely to cost a trick as the side suit queen can sometimes be ruffed out.

 

On the other hand, playing transfers declarer will usually have a range of 15-17 when dummy may have 0-7 or 10-15, so it is easier for the defenders to place the honours when the more narrowly defined hand is declarer. The same sometimes applies to distributions although that obviously depends on methods. Smolen, for example, discloses less about the nt opener's shape than about partner's shape.

 

It would be nice to see some single dummy analysis, as this is a very important issue for system design: under what circumstances should system designers worry about right-siding? But this is obviously quite difficult because we would have to compare some specific systems (say standard transfers vs 2-way stayman) to be able to make assumptions about what information has been leaked.

Being an old timer, one thing I always missed was being able to force and suggest slam with direct 3-bids. I don't know if it is mirage or true but I felt like I got better information from partner when it started 1N-P-3H or the like.

[mycroft]

I know several people who still play that, with a sound transfer system. My current system sort of handles it with two-step 3♥ KC [*] leading into cuebidding, which is almost, but not quite, the same as the Suit Set Slam Try I play with others.

 

Of course, I'm famous for saying "I don't care what we agree on for 3 bids over 1NT, they'll never come up - unless we don't agree on something." That's for pickups, but still something to think about. The gain you get from X system over Y involving 3 bids over 1NT likely will be doubled by working on anything else for that time.

 

[*] There's a bailout: 3NT is a skipped step and says "I'm Jx or worse in trump". Also very helpful, especially if the set trump suit is a minor.

[pescetom]

Here, having the balanced hand declare may offer an additional advantage since if dummy has a singleton in the suit being led, declarer may have an honour to protect. On the other hand, with Hx opposite Hxx it may be better to have the 3-card suit in dummy. So I would expect it not to make any difference in your sims. Maybe Pavlicek did something wrong, or maybe your results are actually consistent with Pavlicek's results, as some of your hands will have a very weak dummy, and it may be important that declarer is balanced and dummy has a 5+ suit.

I suspect that smerriman hit the nail on the head and that many of the gains are from situations where all four suits lose a trick leading into the strong hand. But yes my inclusion of very weak dummy may be relevant too: I will try setting a range which is easy enough to do.

If I find time I will also try the more complex simulation of 4-4 fits discovered through Stayman: these are usually rightsiding in a 5-card major system, and more frequent (and less varied in strength) than 5+ card transfers too.

 

 

 

On the other hand, playing transfers declarer will usually have a range of 15-17 when dummy may have 0-7 or 10-15, so it is easier for the defenders to place the honours when the more narrowly defined hand is declarer. The same sometimes applies to distributions although that obviously depends on methods. Smolen, for example, discloses less about the nt opener's shape than about partner's shape.

Revealing more precise HCP information is certainly my main reserve about opening more hands in NT than others do. But it helps responder during the auction, so it's only fair it helps the opponents later. They don't get as much information about declarer's hand however and my Stayman is obsessive about concealment.

[Winstonm]

I suspect that smerriman hit the nail on the head and that many of the gains are from situations where all four suits lose a trick leading into the strong hand. But yes my inclusion of very weak dummy may be relevant too: I will try setting a range which is easy enough to do.

If I find time I will also try the more complex simulation of 4-4 fits discovered through Stayman: these are usually rightsiding in a 5-card major system, and more frequent (and less varied in strength) than 5+ card transfers too.

 

 

 

 

Revealing more precise HCP information is certainly my main reserve about opening more hands in NT than others do. But it helps responder during the auction, so it's only fair it helps the opponents later. They don't get as much information about declarer's hand however and my Stayman is obsessive about concealment.

 

My guess based on nothing but personal experience is that game strength hands gain from transfers while slam-try strength hands and weak hands are better served concealing their strength/weaknesses.

[pescetom]

Here, having the balanced hand declare may offer an additional advantage since if dummy has a singleton in the suit being led, declarer may have an honour to protect. On the other hand, with Hx opposite Hxx it may be better to have the 3-card suit in dummy. So I would expect it not to make any difference in your sims. Maybe Pavlicek did something wrong, or maybe your results are actually consistent with Pavlicek's results, as some of your hands will have a very weak dummy, and it may be important that declarer is balanced and dummy has a 5+ suit.

I modified the script (see below) to identify a weak dummy (0-7) or not (8+) to see how that was impacting things.

As I expected, the weak dummy benefits considerably more (6.4% gain, 1.8% loss) than an invitational+ dummy (5.3% gain, 3.5% loss).

I trust Winstonm is taking note :)

To be fair, the gain is against playing the weak hand in spades and not against passing 1NT; if anyone has doubts I can add that option.

 

 

 

#
# Does transfer to the stronger hand gain tricks?
#
produce 10000
N1NT = shape(north, any 4333 + any 4423 + any 5332) and hcp(north)>=15 and hcp(north)<=17
S5S = spades(south)>=5
Sweak = hcp(south)<8
Ntricks = tricks(north,spades)
Stricks = tricks(south,spades)
delta = Ntricks - Stricks
action frequency "of spades tricks" (Ntricks,7,13), frequency "of extra spades tricks due to transfer" (delta,-2,2)
condition N1NT and S5S and (not Sweak)

[helene_t]

I find it intriguing that you get such a big strong-declarer advantage compared to Pavlicek. Pavlicek found, for example, that when the points are 16-9 and you have a major suit fit, the average number of tricks is 10.33 from the strong hand and 10.32 from the weak hand.

 

I suspect it's more about balanced vs unbalanced than strong vs weak. Which would justify that people tend to play transfers even if they pay weak NT, while transfer responses to strong club openings is more something for geeks.

 

http://www.rpbridge.net/8j21.htm

[pilowsky]

I wonder if there is a problem related to the amount of variance accounted for by the difference.

When you say that there is a "considerable" advantage, you look at the Loss versus Gain data without considering the total.

In your simulation (which btw also works for other 1NT trump strengths and shapes), the proportion of the effect compared to the total number of tricks seems very small (below the .05 level).

It seems that rather than being a "considerable" effect, it is, in fact, a trivial effect that is likely swamped by the many other factors that come into play, including:

• Opposition bidding
• Quality of play by Declarer
• Quality of play by the opposition

Obviously, an advantage is better than a disadvantage.

But, with every upside comes the memory load needed to get it right. This "memory load" is part of the denominator that might wash away any advantage.

 

It is also noteworthy that to generate 1000 hands, the program had to generate more than 1 million hands.

Does this mean that the real denominator is several orders of magnitude higher in getting it right?

 

Modern 1NT warfare seems to have expanded well beyond the parameters that you have set (upgraded 14's and downgraded 18's) very misshapen hands (as those of us that play against robots know).

 

I ran your code with different HCP parameters and the results were rather similar:

 

N1NT = shape(north, any 4333 + any 4423 + any 5332) and hcp(north)>=7 and hcp(north)<=10

Frequency of spades tricks:
Low 		434
   8 		195
   9 		171
  10 		115
  11      	58
  12      	21
  13   		6
Frequency of extra spades tricks due to transfer:
  -2   		0
  -1      	27
   0 		945
   1      	27
   2   		1
Generated 36760 hands
Produced 1000 hands
Initial random seed 1618393724
Time needed   74.057 sec

 

 

 

N1NT = shape(north, any 4333 + any 4423 + any 5332) and hcp(north)>=14 and hcp(north)<=18

Frequency of spades tricks:
Low 		112
   8 		162
   9 		186
  10 		222
  11 		201
  12 		104
  13      	13
Frequency of extra spades tricks due to transfer:
Low   		1
  -2   		0
  -1      	26
   0 		929
   1      	44
   2   		0
Generated 71060 hands
Produced 1000 hands
Initial random seed 1618393854
Time needed   59.343 sec

 

 

N1NT = shape(north, any 4333 + any 4423 + any 5332) and hcp(north)>=11 and hcp(north)<=14

Frequency of spades tricks:
Low 		234
   8 		187
   9 		220
  10 		177
  11 		122
  12      	47
  13      	13
Frequency of extra spades tricks due to transfer:
  -2   		0
  -1      	27
   0 		942
   1      	31
   2   		0
Generated 44029 hands
Produced 1000 hands
Initial random seed 1618393419
Time needed   65.772 sec

[smerriman]

I find it intriguing that you get such a big strong-declarer advantage compared to Pavlicek. Pavlicek found, for example, that when the points are 16-9 and you have a major suit fit, the average number of tricks is 10.33 from the strong hand and 10.32 from the weak hand.

I think you misread the table. It's 10.33 from the hand with more HCP, and 10.32 from the more balanced hand - not the weaker hand. The fixed result is 10.32, so the weak result should be 10.31, resulting in a difference of about 0.02 tricks per hand (plus or minus quite a lot due to the rounding to 2dp).

 

Pescetom's original table averages out to 0.03 tricks per hand, so isn't "big" in comparison (and of course, the original sim wasn't 16-9 with a spade fit).

[helene_t]

I think you misread the table. It's 10.33 from the hand with more HCP, and 10.32 from the more balanced hand - not the weaker hand. The fixed result is 10.32, so the weak result should be 10.31, resulting in a difference of about 0.02 tricks per hand (plus or minus quite a lot due to the rounding to 2dp).

 

Pescetom's original table averages out to 0.03 tricks per hand, so isn't "big" in comparison (and of course, the original sim wasn't 16-9 with a spade fit).

Thanks, you are right of course.

 

If we look at the 21 HCP table, the difference between fixed and stronger is 0.01-0.02, corresponding to a difference of 0.02-0.04 between stronger and weaker declarer for the 15-6 to 17-4 HCP splits. Pescetom got a 4.6% difference for the less-than-invitational dummy's so it's actually roughly the same as Pavlicek's results.

[pescetom]

I wonder if there is a problem related to the amount of variance accounted for by the difference.

When you say that there is a "considerable" advantage, you look at the Loss versus Gain data without considering the total.

In your simulation (which btw also works for other 1NT trump strengths and shapes), the proportion of the effect compared to the total number of tricks seems very small (below the .05 level).

The only thing I said was "considerable" is the proportional difference between the benefit with a weak dummy (4.6% net gain) and the benefit with an invitational+ dummy (1.8% net gain).

[pescetom]

I find it intriguing that you get such a big strong-declarer advantage compared to Pavlicek. Pavlicek found, for example, that when the points are 16-9 and you have a major suit fit, the average number of tricks is 10.33 from the strong hand and 10.32 from the weak hand.

 

I suspect it's more about balanced vs unbalanced than strong vs weak. Which would justify that people tend to play transfers even if they pay weak NT, while transfer responses to strong club openings is more something for geeks.

 

I suspect it's more about the combination of strong declarer plus weak responder (so weak NT transfers may indeed be unjustified double-dummy).

As pilowsky pointed out, lowering the NT range of my script to 11-14 almost balances the gain/loss (and much the same for a traditional 12-14).

Whereas a quick run with an NT range of 20-22 lost 2.6% of times and gained 7.5%.

[foobar]

I hacked the rough script below to see if a simple major suit transfer was gaining/costing tricks.

It assigns North an orthodox 15-17 1NT and assumes that one of NS will declare and play in spades any time South has 5+ spades.

 

My conjecture is that with weak (say 4-7 HCPs) flat hands with poor suit (9xxxx for example), it's better to let opener play in 1N. It might be interesting to see what your simulation suggests in that particular situation.

[pilowsky]

Something that may be worth considering is the opportunity cost of insisting on "right-siding" the contract.

Particularly if the advantage is - on average - within the error of player ability.

By using transfers it is now impossible for a weak responder to bid 2♣ or 2♦ as natural not forcing.

Considering the Stayman structure as a whole, would it be feasible to play all 2 level responses to a 1NT opening as weak with 5 cards in the suit?

 

A Stayman-like game-forcing auction would then start at the 3-level:

1NT - P - 2anything = < than 8 HCP and 5 in the suit.

1NT - P - 3♣ = > 8 HCP and whatever Stayman you want to play AND game forcing.

1NT - P - 3♦ = ♥ GF

1NT - P - 3♥ = ♠ GF

1NT - P - 3♠ = [minors] GF

etc

[hrothgar]

 

Considering the Stayman structure as a whole, would it be feasible to play all 2 level responses to a 1NT opening as weak with 5 cards in the suit?

 

 

I sincerely doubt that this is playable.

 

Consider what you are giving up

 

1. You can't show a five card major at the two level and then follow up with a three level bid

2. You can't make a game invite belong the 2NT level

2. You lose out on Garbage Stayman or the equivalent

 

Some people like to play 2 way Stayman (using 2♣ and 2♦ as G/I and GFing bids. This allows them to use 2M as NNF.

 

If you go further back in time, you have structures like Gladiator that uses 2♣ as a puppet to 2♦

 

However, I have never heard of anything like what you are suggesting

[pescetom]

 

 

If you go further back in time, you have structures like Gladiator that uses 2♣ as a puppet to 2♦

One can also play 2♧ as a marionette to 2◇, to be broken only when opener holds a 5-card major (as in early 'Puppet' Stayman). This offers a fairly sure 2◇ even if responder lacks 3 card support for one or both majors.