When to avoid stayman

[pclayton]

Sorry about the bad data/conclusions. This time I have done:

 

south 4(432) with 15 HCP

north 4333 with 10 HCP

1000 deals

Tim: Can you try it with a 4333 and 12 HCP? 14?

 

Thanks.

[TimG]

I don't doubt that the 4-4 fit will often produce more tricks than NT. However, it is also very true that when you do not find a 4-4 fit, by providing distributional information to the opponents you will have given up some tricks.

 

Furthermore, even when you find the 4-4 fit, the opponents will have more information than if the auction had gone 1NT-3NT. So, even with the large plus for playing in the 4-4 fit (as clearly demonstrated by the simulation), there will also be a minus in that the defenders will have more information.

I would not be surprised if the extra information provided to the opponents is more significant than the 45% v 55% advantage of the 44 fit. But, the value of the extra information is difficult to quantify and I would not want to leap to any conclusions...

[TimG]

Sorry about the bad data/conclusions.  This time I have done:

 

south 4(432) with 15 HCP

north 4333 with 10 HCP

1000 deals

Tim: Can you try it with a 4333 and 12 HCP? 14?

south 4(432) with 15 HCP

north 4333 with 12 HCP

757 deals

 

tricks in spades: 10.09

tricks in no trump: 9.36

difference: 0.73 tricks

matchpoint for 3N: 46.2%

 

south 4(432) with 15 HCP

north 4333 with 14 HCP

384 deals

 

tricks in spades: 10.74

tricks in no trump: 10.28

difference: 0.46 tricks

matchpoint for 3N: 49.1%

 

Decreasing sample size is because the dealer was set to stop generating after 100,000,000 even if 1000 matches were not found. Maybe I'll up the number of generated hands to increase the sample to 1000 deals, but maybe not...

[pclayton]

Pretty amazing stuff here. Have we been doing it wrong all these years?

 

Another thought: If a 4=3=3=3 10 count gets you 8.38 tricks in NT on average, perhaps we should be inviting on hands like this. I realize this is DD, but it certainly tells me that a 4=3=3=3 nine count is definitely an invite.

[han]

What if declarer is allowed to be any of 4432, 4333 or 5332? In other words, if you do not have the methods to distinguish between those hands, is it then double-dummy worth to bid Stayman when you are 4333?

 

I do think that in practice going through Stayman on hands where you end up in 3NT will often cost. Perhaps the worst possible auction is 1NT-2C-2S-(choice of games)-3NT (I'm 4333) - pass (me too!). Of course auctions like 1NT - 2C - 2H - 3NT - pass will also do badly compared to 1NT-3NT.

[han]

Another thought: If a 4=3=3=3 10 count gets you 8.38 tricks in NT on average, perhaps we should be inviting on hands like this. I realize this is DD, but it certainly tells me that a 4=3=3=3 nine count is definitely an invite.

Is inviting with a 4333 9-count at matchpoints new? I guess I've been underbidding as usual. For the 10-count if you seriously consider inviting then I think you don't give enough weight to the double dummy factor.

 

By the way Tim, even for a fair double dummy comparison you should consider all 15-17 hands for opener, not just 15-counts. One would expect the advantage of 4M compared to 3NT to be smaller when opener has 17, just like the advantage is smaller when responder is stronger.

 

I think you made a typo for the 14-count numbers, 9.24 should be 10.24?

[awm]

I didn't make my comments up and happen to get lucky with the simulation. This stuff about bidding 3NT on 4333 at IMPs and trying for 4M at MPs is a known fact that Ron Klinger wrote about quite some time ago (and I don't think he discovered it either). It has also been discussed on these forums before.

 

It's true that there is some effect from giving opponents more information in the bidding. But I think a lot of people tend to overrate the amount of difference this actually makes. I have seen plenty of hands where the stayman auction talks opening leader out of his normal major suit lead and his normal lead actually beats the contract. In any case declarer's advantage over double-dummy play and defense hovers around 0.1 tricks from everything I've seen. Likely it's a bit more after an uninformative auction and a bit less after an informative auction but I doubt it's going to ever be comparable to the 0.7-0.9 average trick differences we are seeing between 3NT and 4M.

 

Admittedly this trick difference tends to vanish when the high card point total reaches 29-30 and at this point bidding stayman becomes bad MP strategy.

 

It may also be worth mentioning that using a 1NT forcing response on major suit limit raise hands potentially gives the opposition even more useful information than bidding stayman over a 1NT opening. Yet people do this routinely even though there is virtually no potential benefit (i.e. they were always going to make their 3-card limit raise no matter what opener's rebid was)!

[TimG]

Pretty amazing stuff here. Have we been doing it wrong all these years?

 

Another thought: If a 4=3=3=3 10 count gets you 8.38 tricks in NT on average, perhaps we should be inviting on hands like this. I realize this is DD, but it certainly tells me that a 4=3=3=3 nine count is definitely an invite.

On an unrevealing auction like 1N-3N, declarer's advantage is probably between 1/4 and 1/2 trick. (This is not just a guess, I did work on this about 10 years ago, even had three others working with me to check the work, but I use "probably" so as not to make any unproven claims.) So, I think blasting with a 4333 10 count is right. Of course, if your range is really 14+-17- instead of 15-17, that may be a different story.

[TimG]

By the way Tim, even for a fair double dummy comparison you should consider all 15-17 hands for opener, not just 15-counts. One would expect the advantage of 4M compared to 3NT to be smaller when opener has 17, just like the advantage is smaller when responder is stronger.

I used 15 to give it a worst case flavor. (When I started, I was sure that skipping Stayman was going to be a winner even given the worst case.) You are right that if opener's range was 15-17, then the choice between 3N and a 44 major suit fit would be closer.

 

I think you made a typo for the 14-count numbers, 9.24 should be 10.24?

Yes, thanks. Now fixed.

[TimG]

In any case declarer's advantage over double-dummy play and defense hovers around 0.1 tricks from everything I've seen. Likely it's a bit more after an uninformative auction and a bit less after an informative auction but I doubt it's going to ever be comparable to the 0.7-0.9 average trick differences we are seeing between 3NT and 4M.

You don't have to overcome the whole 0.7-0.9 difference, just about half of it -- as we saw with 15 opposite 14, the trick difference was about 1/2 and the matchpoint result was about 50/50. And, that's if you only consider the hands with a 44 fit and not the hands were one investigates and comes up empty.

[MFA]

Nice thread!

 

My style is to blast 3NT with any 4333. Not because I expect to gain from playing 3NT with a 4-4, but because I expect to lose so little doing so with 4333 that I earn more from not revealing anything when we don't have a fit.

 

But I'm surely ready to learn.

[awm]

I've seen pretty convincing information (from many sources) indicating that declarer's double dummy advantage (on average over all hands) is about 0.1 tricks.

 

Uninformative auctions (1NT-3NT, 1♠-3♠-4♠, 1NT-PASS, etc) are pretty common. If declarer's advantage on these hands was 0.5 tricks, it would mean that the "informative" auctions outnumber the "uninformative" auctions by at least 4:1, or that declarer is actually at a double-dummy disadvantage on a mildly informative auction. In any case either would imply that while declarer's "mean tricks over double dummy" is 0.1, his "median tricks over double dummy" would be substantially less, likely actually negative. This situation seems fairly unbelievable to me.

 

So I suspect that the advantage is much closer to TimG's 0.25, if not substantially lower.

 

It may also be worth noting that declarer could easily have a large advantage in the auction 1NT-3NT and actually have almost as large an advantage in the auction 1NT-2♣-2♦-3NT. In other words, even if declarer has a 0.4 trick advantage in the first case, if he also has a 0.35 trick advantage in the second case then the difference is not all that significant.

 

As a last point, if informative auctions really create a difference of something like 1/4 trick per board, why haven't any of the expert pairs who use forcing notrump on three-card limit raises caught on yet? Seems like a pretty big price to pay for basically no benefit (and it actually makes your slam bidding worse when opener jump shifts too).

[TimG]

I've seen pretty convincing information (from many sources) indicating that declarer's double dummy advantage (on average over all hands) is about 0.1 tricks.

Brief summary of the work I did some 10 years ago:

 

I used a few million OKBridge deals. I dealt primarily with straightforward NT auctions. 1N-p-p-p held the biggest declarer's advantage (DA) at almost 0.5 tricks. 1N-3N had less DA, but it was not that far behind.

 

Some high level contracts had a near zero DA, but I never found any that had negative DA.

 

The advantage was often entirely the opening leader's disadvantage -- often there was negative DA after the opening lead was made, that is defenders did better relative to double dummy result after the opening lead.

 

On deals where there were some minimum number of both 1N-3N and 1N-2C-2x-3N auctions, there was a statistically significant difference in result relative to double dummy (with declarer doing better in the 1N-3N auction). This even held true for 1N-3N vs 1N-2N-3N.

 

The hands studied were OKbridge hands without any filter for strength of players. So, you may rightly question whether these results reflect what you would expect in an "expert game".