[awm]

We are considering switching from 2/1 forcing one round to forcing through 2NT. I'm interested in the following comparison:

Opener: 5143 or 5242 or 5341 with 10-11 hcp
Responder: 2434 or 2425 or 2335 with 12-13 hcp

Which is better and by how much: 2S played by opener, or 2NT played by responder? I'm interested in both IMP and MP scoring.

We believe that 2NT is an easier contract to declare single-dummy (because 2S often requires deciding whether to draw trumps or try to elope with the spades, and which is best is frequently determined by the trump break) and also that playing "forcing through 2NT" will improve our slam auctions considerably... so we want to switch if playing 2NT in these partials is "not much worse" double dummy.

We don't want to play 2/1 GF because our light-opening style messes with the frequencies and because we find having two different invitational sequences (one via 2/1 and one via 1NT) to be really valuable given our fairly wide range and light style.

Thanks,
Adam

[FrancesHinden]

I realise this isn't answering the question, but our empirical experience is that overall we would rather play 2M than 1NT. 2NT is a truly horrible spot.

[Fluffy]

I've played forcing to 2NT all my life before switching to 2/1, I really don't understand bidding the other way. 2NT ain't great but I don't think we play it more than 3 times/year, obviously this happens in our context where it will only be landed when responder has exactly 11, opener is minimum and no fit is found. Yours will be wider.

For us 1♠-2x-2y-2♠ is not forcing, but I wouldn't say it is mandatory.

[Siegmund]

If either hand were weaker, 2M rather than notrump would be a no-contest favorite. 23 HCP however is very close to the 50-50 spot for being able to take 8 tricks in notrump which makes this a question worth asking.

1000 deals under the requested conditions:

Spades won 8520 tricks. Notrump won 7938 tricks.
At matchpoints, spades won 531 times, tied 52 times, lost 417 times. More than enough for spades to be statistically significantly better (scoring 55.7±3% vs. pairs in 2NT.)

I confess I forgot to have the sim calculate imp differences, and it takes awhile to re-run, but I can't imagine the result being much different, since the swings are 1, 2, and 4 imps in either direction, according to whether one contract makes an overtrick, avoids an undertrick, or makes when the other fails, no big lopsidedness like in a 3NT vs 3m sim where the size of the game bonus matters.

Ultimately, no surprise. It's incredibly hard to construct any set of hands where 2NT isn't a disaster.

[lowerline]

I don't play 2/1 GF either, but like Frances I also prefer playing 2M over 2nt. In my system it is even impossible to end in 2nt after a 2/1 by making 2nt GF.

Steven

[rhm]

 Siegmund, on 2013-August-03, 22:30, said:

If either hand were weaker, 2M rather than notrump would be a no-contest favorite. 23 HCP however is very close to the 50-50 spot for being able to take 8 tricks in notrump which makes this a question worth asking.

1000 deals under the requested conditions:

Spades won 8520 tricks. Notrump won 7938 tricks.
At matchpoints, spades won 531 times, tied 52 times, lost 417 times. More than enough for spades to be statistically significantly better (scoring 55.7±3% vs. pairs in 2NT.)

I confess I forgot to have the sim calculate imp differences, and it takes awhile to re-run, but I can't imagine the result being much different, since the swings are 1, 2, and 4 imps in either direction, according to whether one contract makes an overtrick, avoids an undertrick, or makes when the other fails, no big lopsidedness like in a 3NT vs 3m sim where the size of the game bonus matters.

Ultimately, no surprise. It's incredibly hard to construct any set of hands where 2NT isn't a disaster.

I doubt that this indicates spades to be better at matchpoints single dummy.
Double dummy results are sometimes tricky to interpret.
Single dummy, declarer enjoys a greater advantage at a notrump partial compared to 2M, mainly due to the opening lead importance at notrumps.
More notrump partials and games are made, which should have gone down than suit contracts.
In my experience you need to have at least an advantage of 0.7 tricks on average more in a major double dummy before you should reject a notrump partial or game.
However, your simulation probably indicates that at IMPs 2M will usually be the safer partial.

The case remains close at matchpoints.

Rainer Herrmann

[rhm]

My simulation gave the following results:

1000 deals:

♠ produced 8.444 tricks on average
notrump produced 7.871 tricks on average

2♠ made 84% of the time
2NT made 64% of the time

when 2♠ was down (156 deals), 2NT made in 35% of the deals.
when 2♠ made (844 deals), 2NT made in 69% of the deals
when 2NT was down (362 deals), 2♠ made on 72% of the deals
when 2NT made, 2♠ made on 91% of the deals.

When 2♠ made a certain number of tricks, 2NT did take the same number of tricks on the majority of the deals.

This all points to the conclusion that 2♠ is the safer contract, but at matchpoints 2NT may have the higher matchpoint expectancy.
Single dummy the results will point even more in direction of notrumps.

By the way, I can not confirm your frequency issue. Your simulation question is not "quick". My software had to try more than 25000 deals on average for every one, which met all your criteria.
These precise circumstances are rare and I would not have sleepless nights over them. You will not get many of them in your remaining lifespan.
Even if you open light, the mildly misfitting invitational 12+ hands (of course there are more of them), where you would like to stop below game, are quite infrequent.
This might be the reason why a lot of pairs play 2/1 and still open light and get rarely hurt at the game level.

Rainer Herrmann

[TylerE]

What %age of the time 2N made would 3N make?

When 2N is making exactly, how often does 3♠ make?

[rhm]

 TylerE, on 2013-August-05, 08:14, said:

What %age of the time 2N made would 3N make?

When 2N is making exactly, how often does 3♠ make?

When 2NT is making exactly (8 tricks), you make double dummy 9 or more tricks in spades 53% of the time.
In 44% of the cases when you make 2NT, you make overtricks that is 3NT as well.
I repeated the simulation (very similar results) for your questions, since I had discarded my data.

Rainer Herrmann

[jogs]

Quote

Opener: 5143 or 5242 or 5341 with 10-11 hcp
Responder: 2434 or 2425 or 2335 with 12-13 hcp

I'm really surprised that DD yielded about 8.5 tricks for these
patterns and point ranges. I was expecting under 8 for any
DD test.

Maybe we should narrow the comparisons. There are 3 patterns
for opener and 3 patterns for responder. That is 9 pattern pairs.
There are two point ranges for each players. A possible 3
point range for total points.

Maybe we should test 5143 11 HCP for opener against 2434
with 12 HCP.

5143 // 2434
Then exchange responder's low diamond and for a low club.
5143 // 2425

How much is the fifth card in the club suit worth?

TIA, jogs

[Siegmund]

Your new tighter conditions make the sim much slower (unless you do a bunch of extra programming to generate a list of suitable opening hands, and then a responding hand for each, rather than just dealing random hands and hoping both hands qualify.)

I show the 5143-2434 hands averaging 8.58 tricks in spades and 7.98 in notrump, and the 5143-2425 hands averaging 8.72 tricks in spades and 8.16 in notrump.
It's a statistically significant difference, but dont put too much faith in the exact size of the spread (0.14±0.11 and 0.18±0.11 tricks), unless you construct a special experiment where you deal two possible responding hands for each opener and see what the difference in numbers of tricks is.

rhm wrote:

Quote

In my experience you need to have at least an advantage of 0.7 tricks on average more in a major double dummy before you should reject a notrump partial or game.

This is quite a surprising claim to me; it's true that defenders make worse opening leads against notrump than against suits -- especially against uninformative auctions like 1NT-3NT -- but my sims have put the cost of a blind lead against 1NT-3NT at most 1/4 of a trick worse than against an informative auction, and I would expect the difference to be a lot smaller than that if we are comparing 1S-2D-2NT and 1S-2D-2S (or two other similar auctions which give almost equally much information about declarer's hand.) Even a tenth of a trick is more of a difference than I would expect.

[rhm]

 Siegmund, on 2013-August-05, 23:44, said:

This is quite a surprising claim to me; it's true that defenders make worse opening leads against notrump than against suits -- especially against uninformative auctions like 1NT-3NT -- but my sims have put the cost of a blind lead against 1NT-3NT at most 1/4 of a trick worse than against an informative auction, and I would expect the difference to be a lot smaller than that if we are comparing 1S-2D-2NT and 1S-2D-2S (or two other similar auctions which give almost equally much information about declarer's hand.) Even a tenth of a trick is more of a difference than I would expect.

At matchpoints a half trick advantage on average in favor of playing in a major compared to notrump is in itself no reason to play the major, since you are as likely to make the same number of tricks in notrump than one trick less.
If you make the same number of tricks you win playing in notrumps, if you make less tricks you win playing the major.
Comparisons between double dummy and actual single dummy plays show that low level notrump contracts score about 0.2 tricks better on average than suit contracts at the same level single dummy than double dummy results suggests. Both do better than double dummy results at low levels, but notrump even more.
This comparison does not distinguish between revealing bidding sequences and less revealing ones. Revealing sequences will make the opening lead and defense more precise in both cases.
For example in 2♠ I would expect a trump lead to be forthcoming more often when that is right after such an informative sequence.
The results I know compare single dummy outcomes from actual play to double dummy ones, separated according to level and whether notrump or suit contracts.
It does fit my subjective experience well. When I analyze my results nowhere do I beat double dummy results more often than at notrumps. My subjective estimate is that a 0.2 tricks notrump advantage is conservative.

Considerations for IMPs are quite different and generally more complex.
At IMPs 3NT is favored even more since you need 2 tricks more in a major before you will be substantially better off at IMPs.
But when it comes comparing 2♠ to 2NT you want the safer contract. Overtricks are of lesser importance.
Notrump contracts tend also to have a larger variance in outcome than suit contracts.
It is not uncommon to make only 6 or 7 tricks with 25 HCP combined and sometimes 9 tricks with 21 HCP combined for reasons undetectable in the bidding.

Rainer Herrmann

[Siegmund]

A half trick for the scoring difference and 0.2 for quality of defense sounds much more reasonable (when both contracts are making: 2S+2 and 2NT+2 is a win for notrump while 2S-1 and 2NT-1 is a tie) -- I never thought of it that way since in my sims I generally directly count the matchpoint wins and losses rather than just looking at the trick differences at the end.

[jogs]

 Siegmund, on 2013-August-05, 23:44, said:

Your new tighter conditions make the sim much slower (unless you do a bunch of extra programming to generate a list of suitable opening hands, and then a responding hand for each, rather than just dealing random hands and hoping both hands qualify.)

I show the 5143-2434 hands averaging 8.58 tricks in spades and 7.98 in notrump, and the 5143-2425 hands averaging 8.72 tricks in spades and 8.16 in notrump.
It's a statistically significant difference, but dont put too much faith in the exact size of the spread (0.14±0.11 and 0.18±0.11 tricks), unless you construct a special experiment where you deal two possible responding hands for each opener and see what the difference in numbers of tricks is.

The fifth club is only worth an extra trick about once every seven boards?
The +0.11 looks like variance. The std dev is more like +0.33.

Is it possible to create a file of the first group of sims? Then write a
program that exchanges cards between South and West. Run the same boards
with the minor change.

[Siegmund]

Quote

The fifth club is only worth an extra trick about once every seven boards?

Yes. (And that is much higher than I expected it to be. In many notrump sims, a 5-card suit is not an asset at all, double-dummy -- the risk of having one suit be shorter costs more than the potential for an extra winner in the long suit.)

Quote

The +0.11 looks like variance. The std dev is more like +0.33.

The ±0.11 is the width of a 95% confidence interval for the value of the fifth club, based on the data I had.
I had 1000 2434 hands, which produced an average of 8.58 tricks with a standard deviation of 1.00 tricks, and 600 2425 hands, which took 8.72 tricks with a standard deviation of 1.10 tricks: 1.96 * sqrt(1.00^2/1000 + 1.10^2/600) ~ 0.107.

Quote

Is it possible to create a file of the first group of sims? Then write a
program that exchanges cards between South and West. Run the same boards
with the minor change.

Possible? Yes. Worth my time to make? No. (I have a number of reusable programs to do various types of analysis quickly - but that isn't one of them.) It would have to be done a bit differently than that, too, rather than always removing the extra club from west.

[jogs]

Opener: 5143 with 11 hcp
Responder: 2434 with 12 hcp
...........
Let's see if I got this right.
* Both Siegmund and Rainer agree that in spades the
DD expected tricks is approximately 8.5.
* Rainer says the difference between the DD results
and actual play in spades is small.

Can we assume that the 6-2 fit yields more expected tricks
than the 4-4 fit?
.................

Our side has exactly 20 HCP. We have a 5-3 fit in our longest
suit fit. Another 8 card fit is allowed as long as it is 5-3 or 4-4.
What is our expected tricks? Without benefit of any study,
my guess was 8 tricks. Is this close(+/- 0.10)? Is it possible
the 5-3 fit yields more expected tricks than the 4-4 fit?