[rogerclee]

IMPs, All White

♠AKQxxx ♥xx ♦xxx ♣xx

1NT-(P)-?

Partner opens a 15-17 NT in first seat. Texas to 4♠ or 3N?

A simulation would be nice, if anyone wants to run one of those.

[655321]

There is a 3rd option which I would choose, transfer then 3NT.
Second choice, 3NT immediately.

[rogerclee]

655321, on Mar 20 2008, 07:19 AM, said:

There is a 3rd option which I would choose, transfer then 3NT.

Oh, yeah, this is pretty logical, and I forgot to include it.

[brianshark]

My guess is texas. But yeah, sim would be interesting.

[whereagles]

A 15-17 has ~4 cover cards. The hand has 7 losers.

Hence, transfer + 4♠. 3NT is a gamble; to make when you need a swing.

[awm]

I would transfer at the two-level and bid 3NT. Obviously nothing is guaranteed to always get you to the best game, but at least this way if partner has 3-4♠ and a small doubleton on the side we are guaranteed to play 4♠ and not 3NT. I'll take my chances that if partner can't ruff anything 3NT is probably better than 4♠.

[jdonn]

I think 4♠ is 100% clear since it's far from guaranteed that the spades will run in 3NT, whereas if they don't break 4♠ is still often making. It would be a more interesting problem with AKQJxx.

[mikeh]

jdonn, on Mar 20 2008, 10:55 AM, said:

I think 4♠ is 100% clear since it's far from guaranteed that the spades will run in 3NT, whereas if they don't break 4♠ is still often making. It would be a more interesting problem with AKQJxx.

Of course, if partner has xx in spades and side suits double stopped (not that improbable given that spade holding) he may be able to afford to duck the 1st spade... and if we have a spade loser, 9 tricks may be easier than 10. I am not saying that your point is entirely wrong, but it isn't as simple as 'will the spades run'.

In fact, in a simulation based on a small sample of 32 hands, both 3N and 4♠ made most of the time, there were only 3 where neither game made, and only 1 where 4♠ made and 3N failed... but there were 7 where 3N made and 4♠ failed... on most (but not all) of them, the key was ducking the 1st round of spades.

The constraints were that S had precisely 2 spades... obviously, if he has 3 the odds of losing a spade trick are far lower.

I didn't simulate hands where opener had 3 or 4 spades.... my expectation would be that hands on which opener held 4 spades will, where it makes a difference, strongly favour 4♠ and that the transfer and 3N route caters to those, and to those where opener has 3 spades and a suitable hand to pull 3N.

[TimG]

My simulation suggests there is approximately one extra trick in spades and that 3N fails more often than 4S (269 to 197 in 1000 hands).

Adding the Jack of Spades improved the chances of both contracts, 3N failed 174 and 4S failed 79 times out of 1000, but seems to further support 4S.

In my opinion, a not insignificant factor is the blind opening lead if responder does not show spades -- it should be easier to defend against 1N-4H-4S than 1N-3N. Double dummy simulation might not be best here. Searching actual results with Bridge Browser and comparing those results to the double dummy results could be quite useful.

[mikeh]

TimG, on Mar 20 2008, 12:05 PM, said:

My simulation suggests there is approximately one extra trick in spades and that 3N fails more often than 4S (269 to 197 in 1000 hands).

Adding the Jack of Spades improved the chances of both contracts, 3N failed 174 and 4S failed 79 times out of 1000, but seems to further support 4S.

In my opinion, a not insignificant factor is the blind opening lead if responder does not show spades -- it should be easier to defend against 1N-4H-4S than 1N-3N. Double dummy simulation might not be best here. Searching actual results with Bridge Browser and comparing those results to the double dummy results could be quite useful.

Your simulation generates hands and then uses a deep finesse or similar double double analysis to predict the outcome?

I don't do this when I simulate hands.. which is why my samples are always small. I strongly believe that double-dummy simulations of outcomes are misleading. For example, opening leader will never give away the 9th trick on opening lead against 3N, and will always hit partner's suit when right to do so, yet in real life will always get these wrong.

DD simulations are fast and convenient, but they mislead rather than illuminate.

Put another way: if we were always facing double dummy defence, we'd all be bidding far more conservatively... even if allowed to double dummy the declarer play.

[han]

I agree with transfer followed by 3NT. I suspect that 3NT will more often be a better spot when partner passes.

[TimG]

mikeh, on Mar 20 2008, 12:19 PM, said:

TimG, on Mar 20 2008, 12:05 PM, said:

My simulation suggests there is approximately one extra trick in spades and that 3N fails more often than 4S (269 to 197 in 1000 hands).

Adding the Jack of Spades improved the chances of both contracts, 3N failed 174 and 4S failed 79 times out of 1000, but seems to further support 4S.

In my opinion, a not insignificant factor is the blind opening lead if responder does not show spades -- it should be easier to defend against 1N-4H-4S than 1N-3N.  Double dummy simulation might not be best here.  Searching actual results with Bridge Browser and comparing those results to the double dummy results could be quite useful.

Your simulation generates hands and then uses a deep finesse or similar double double analysis to predict the outcome?

Yes, my simulation used a double dummy analysis.

Quote

For example, opening leader will never give away the 9th trick on opening lead against 3N, and will always hit partner's suit when right to do so, yet in real life will always get these wrong.

They won't "always" get these wrong in real life.

Quote

I don't do this when I simulate hands.. which is why my samples are always small.

A 32 hand sample is likely statistically insignificant.

Quote

I strongly believe that double-dummy simulations of outcomes are misleading.  DD simulations are fast and convenient, but they mislead rather than illuminate.

I think that if you understand the limitations of a double dummy simulation, they can be useful (and fast and convenient). I also suspect that a double dummy simulation can produce very accurate results in some cases. For instance, in the case at hand, we can get a good idea of how the presence of the Jack of Spades changes things. Even if we don't believe in the exact number of tricks predicted by the double dummy analysis, the difference in tricks with and without the Jack should be pretty reliable.

Note that in my post I mentioned the blind opening lead in an auction like 1N-3N and suggested searching actual results. I did not mean to suggest that the double dummy simulation was definitive.

I have compared double dummy results to table results for tens of thousands of hands. Declarer's Advantage does exist. And, it does change depending on the information passed in the auction and the level of the contract.

Quote

Put another way: if we were always facing double dummy defence, we'd all be bidding far more conservatively... even if allowed to double dummy the declarer play.

You are right that Declarer's Advantage would better be referred to as Opening Leader's Disadvantage. When I looked into the matter, I found that after the opening lead, the defense held the advantage (relative to double dummy).

[mikeh]

TimG, on Mar 20 2008, 02:30 PM, said:

Quote

For example, opening leader will never give away the 9th trick on opening lead against 3N, and will always hit partner's suit when right to do so, yet in real life will always get these wrong.

They won't "always" get these wrong in real life.

Quote

I don't do this when I simulate hands.. which is why my samples are always small.

A 32 hand sample is likely statistically insignificant.

What I meant by my comment on leads applies to situations in which there is a clear book lead. For example, how often would one lead from Kx in a suit, into the strong notrump opener, while holding QJ109 in a side suit? Well, when it is double dummy to do so, deep finesse will have the K led. Other than Katz-Cohen, in their heyday, (almost)nobody finds leads like this at the table.

As for the sample size, I now extended it to just over 100 hands. In total, 4♠ made while 3N was going down on normal play 6 times while 3N made while 4♠ failed 10 times.

I agree that the sample size is still small, but I think the simulation offers strong support for, at the least, offering 3N as a place to play.

Now, if I were evaluating the relative merits of blasting either game, I would very strongly suspect that 4♠ is significantly the better choice. But involving partner, as so often is the case, will result in missing many of the hands on which the 3N blasters fail.

[jdonn]

Jlall, on Mar 20 2008, 03:35 PM, said:

I don't understand transfer then 3N, we will find 3N opposite a doubleton spade and 4S opposite 3+? Why do we want to play 3N opposite a doubleton spade?

typ

[mikeh]

Jlall, on Mar 20 2008, 03:35 PM, said:

I don't understand transfer then 3N, we will find 3N opposite a doubleton spade and 4S opposite 3+? Why do we want to play 3N opposite a doubleton spade?

because, based on a simulation of 100+ hands, without assumptions as to double dummy defence (or play), there appears to be a slight advantage to 3N: it makes when 4♠ fails a few more times than the reverse. The sample is small to the point that it is probably more accurate to say that 3N is as good as or maybe slightly better than 4♠.

Bear in mind that 3N will usually be in trouble only when spades break poorly.. which means that we start life with a sure spade loser. There were a couple of hands in the sample in which spades broke 5-0, and 3N still made, while 4♠ had no play.

[awm]

It seems pretty clear that texas is better than bidding 3NT. Tim's simulation very much supports this, and it's easy to construct hands where opener has a weak doubleton somewhere and 3-4♠ and 4♠ is cold while 3NT fails.

Comparing transfer...3NT to texas is trickier. The hands where we really want to play in 3NT are hands where partner has no ruffing value (so spades are unlikely to produce an extra trick) and has reasonable stoppers in the other three suits (so we're not likely to get some suit run on us in notrump). It is fairly easy to construct such hands where 3NT is quite good and 4♠ is lousy. The problem is that partner's decision of whether to pass 3NT or bid 4♠ is not exactly based on this criterion.

However, it does seem that if partner has only two spades, he is less likely to be able to ruff anything. He is also more likely to have the other suits stopped, since he (probably) doesn't have doubleton in any of them and length (i.e. Jxxx) can be an effective stopper in any case. If partner is 4333 and passes 3NT it is also high probability to be right (now spades almost surely run). So transfer and bid 3NT gets us to 3NT when partner has 4333 or two spades, and 4♠ otherwise. My bet is that 3NT is usually (okay not always, but usually) better when partner has 4333 or two spades. Mikeh's simulation seems to agree.

[Cascade]

Texas is clear 3NT maybe horrid opposite two small and a bad break but we might still make 4♠ with a trump loser.

Justin is exactly right (as he often is) - why do we want to play 3NT opposite two small?

I'll do a simulation too.

[TimG]

mikeh, on Mar 20 2008, 03:06 PM, said:

What I meant by my comment on leads applies to situations in which there is a clear book lead. For example, how often would one lead from Kx in a suit, into the strong notrump opener, while holding QJ109 in a side suit? Well, when it is double dummy to do so, deep finesse will have the K led. Other than Katz-Cohen, in their heyday, (almost)nobody finds leads like this at the table.

Correct. But the book lead is often the best lead on a double dummy basis (or at least equally as good as others). So, this will not affect all hands, or even a majority, I would guess. (Would be something worth studying, though.)

[TimG]

mikeh, on Mar 20 2008, 03:06 PM, said:

As for the sample size, I now extended it to just over 100 hands. In total, 4♠ made while 3N was going down on normal play 6 times while 3N made while 4♠ failed 10 times.

I think this highlights the small sample size issue.

After 32 deals, you found 1 where 4S made and 3NT did not and 7 where 3NT made and 4S did not.

Over the next 70+ deals, you found 5 where 4S made and 3NT did not and 3 where 3N made and 4S did not.

I don't think that the neighborhood of 100 deals is significant when you're taking about only 15% of the deals matching the criteria (of one contract failing while the other makes) and the score is 10 to 6 on those deals.

I want to emphasize that I am not claiming that the double dummy approach gives a definitive answer here. I'm simply saying that a sample size of 16 hands isn't nearly enough to draw any firm conclusions.