| Board 20 West Deals Both Vul |
| ||||||||||||||||||||||||||||||
|
|
| |||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||
S 3♠; NS 2N; N 2♠; NS 1♥; EW 1♦; NS 1♣; Par +140
| West | North | East | South |
| Pass | Pass | Pass | 1 ♣ |
| Pass | 1 ♠ | Pass | 2 NT |
| Pass | 3 ♣ | Pass | 3 ♠ |
| Pass | 4 ♠ | All pass |
| Lead: ♣ 5 |
3 ♣=checkback stayman
At another table: p p p 1 ♣ p 1 ♦ p 1nt p 2 ♥ p 2 ♠ p 2nt p 3 ♠ ap. Lead= ♥ 7
NS sequence is standard ACOL with checkback stayman. Is there any logic to North passing 2NT or 3♠?
ReplyDeleteAt the table instinct said to pass but "logic" said I have enough values to press on.
ReplyDeleteFirst, a look at what DDA says about the matter.
a) with that specific north hand passing 2NT is far better than 3NT (8.1 av tricks, makes 9 tricks only 35% of time, MPs 58% for 2NT);
b) with the additional information after checkback that south has 3 spades and less than 4 hearts, passing 3S is a lot better than bidding 4S (gets 74% of MPs)
So DDA confirmed instinct not the "logic". Question is why? DDA will test this.
Ok the results are in: I started with 10 billion random hands, then selected suitable hands where north has 7hcp and specifically 5323 shape as per the actual hand, opposite an ACOL rebid 2NT hand (which can include some 5422 shapes) and 3 spades and less than 4 hearts, with >3 clubs (as per the opening bid).
ReplyDeleteArmed with this info, game is a good bet, makes on ~68% of the hands (which is what my 'logic' told me), so clearly there is some major stuff happening that reduces the success to ~26% on the actual hand (which is what my 'instinct' told me). Now we start the process of elimination
How do we reconcile the 26% actual with the 68% generic? A likely candidate is lack of trump honours (other work especially single dummy analysis hi-lites the importance of this). Reducing north to only those that have <1 of the top 3 trumps reduces the generic to 58%, a 10% swing, making it ,<1 of the top 4 trumps reduces the generic to 55%, a 3% swing. What else? No aces. Reduces the generic to 49%, a 6% swing. What else? No tens. Reduces the generic to 39%
ReplyDeleteI have gone a long way to explaining the difference with a residual between the actual and the generic with ~13% still to be explained. However I have run out of hands to do sensible analysis of! The last run was of slightly less than 50 hands!
Like Hesitator, as South, I was worried about 4S as a contract even though I knew we had a 5-3 fit. Still, I thought we might just creep in with 3NT. And so it turned out. Even if DDA says it shouldn't make, the opponents cannot see all the hands and any defender who doesn't make a good few mistakes every session they play is either brilliant or self-deceiving. In this case, West didn't make a mistake with her lead (a heart) but East held up for a round (maybe trying to reduce entries to dummy) and gave me the extra breathing space I needed to generate nine tricks.
ReplyDeleteI took a look at the opening lead bias when I first started doing DDA. There is slight bias in favour of actual declarer, overall of ~0.1 average tricks, but starts at higher at the 1 level and fades through to the 6 level where it reverses slightly in favour of DDA. At world champ level 3NT was 0.11 av tricks in favour of declarer.
ReplyDeletePavlicek did an interesting test where he compared DDA pre vs post actual opening lead. The difference was ~5% and 0.15-0.2 av tricks in favour of post opening lead DDA. WRT this hand that would adjust the % up from 35% to 40% and 8.1 tricks to 8.3 tricks, still not enough to justify bidding 3NT ... unless of course u believe that our club play is particularly bad!
I went back and did a comparison of 3NT vs 4S with south having 3 spades and <4 hearts (although if you were planning to bid 3NT anyhow you wouldn't bother to use checkback). Very close, 4S wins by 51.6% of MPs to 48.4% of 3NT.
ReplyDeleteAren't all these analyses based on being able to see all four hands? Surely, with 26 points and two balanced hands with a 5-3 major fit, everyone in the room should be in game, and most of the time they are going to make it. Three finesses fail. That's what makes it difficult in this case.
ReplyDeleteYes and no. To start with I used the actual North hand and then randomly deal 5000 hands for the other 3 seats with the constraint being that south has 18-19hcp and balanced, with 4+clubs (and after checkback 3 spades,<4hearts). DDA is then done on each of these 5000 deals. This showed that it wasn't a good idea to go to game, with 3NT making only 35% of time, and 4S after checkback found a 5-3 fit making only 26% of time, ie there is something pretty ugly about this hand that makes the odds defy the normal expectation that game should make.
ReplyDeleteThen I randomly dealt 5000 hands for all 4 seats with instead of the actual north hand, north having 7hcp and the specific 5323 shape (and south as before). Then DDA was done on all 5000 of these deals. This showed that game was a good bet -68% (ie confirms the normal expectation).
From there I tried to reconcile the difference between the actual 26% and the theoretical 68%, by constraining what specific high cards north could have. So in order, I did no akq of trumps, no j trumps, no aces, no tens. All of these had a cumulative impact reducing the theoretical down to 39%, leaving a net 13% unexplained (which I suspect may be to do with unprotected honours in the doubleton suit). HTH
Maybe we all need to use a different method of hand evaluation from the standard ones. HCP and LTC both suggest game. Any ideas :-)
ReplyDeleteWell, now that u mention it....Wayne and I started doing some single dummy analysis. Essentially getting GIB robots to both bid and play hands and then analysing the results. This of course was very time consuming as each robot for each trick does a little DDA to determine what to do. Initially got 10k and did some analysis, which showed the value of trump honours vs side suit honours, and the value of additional small trumps, etc. I ran another 10K but haven't analysed (ran means like set the computer running and come back in two weeks!). Apart from that, u get 'learnings' from DDA stuff. Usually 1hcp shifts the dial about 15% or so, there are some things that don't shift the dial as much but combined they do. So eg the features I chose to test above weren't random, but I knew from prior stuff that they shifted the dial but only a bit, but combined it was about 2hcp - I didn't expect it to be so much. So what I try to do on marginal hands is to assess the +ve vs -ve "features". Here this hand had several negative features but I wasn't really sure it fell into the marginal camp.
ReplyDeleteGood luck with that. Is balance between the hands a major factor - eg (13-13 good, 19-7 bad)?
ReplyDeleteearly on I did some stuff on this. Shape (4333 is a 5% negative feature) and hcp split. Split wasn't important until after 18/7. 18/7 was 58% to make 3NT (both balanced and no major fit) (little different to 13/12) but 20/5 was only 50%, and 23/2 was only 35%. Same pattern when using 24hcp combined.
ReplyDelete