[whereagles]

I meant "weak" as in no HHxx, no intermediates.

[chasetb]

 the hog, on 2014-June-02, 06:13, said:

4 to the q is not a weak suit, Nuno.

Funny you should say that. In Why You Lose At Bridge, Mrs. Guggenheim opens 1♥ on Q975 in a 4-card Majors system and 16-18 HCP 1NT, and S.J. Simon, saying only Experts should pull that off. Mind you, Marston would qualify, but for the rest of us it doesn't. I would have led a Spade for sure at MPs (Bird's book of leading versus NT and our hand strength convinced me) and would be choosing between a Spade and a Diamond at IMPs.

EDIT - I found my book - it was a 2443 hand with 15 HCP. Nowadays partner would likely transfer to the 5-2 Spade fit, but RHO holds ♠ KQTxx. Of course, you have a 4-4 Diamond fit with KJxx opposite QTxx, and it splits xxx-Ax

[rhm]

 BillPatch, on 2014-May-30, 09:16, said:

I ran a short double dummy simulation of 100 hands using Jack. The results:
IMPs Contract set with standard low card lead:
♠ = 11%
♥ = 12
♦ = 16
♣ = 7
Best lead at IMPS is a ♦.
Defensive tricks per hand:
♠ = 2.90
♥ = 2.70
♦ = 2.77
♣ = 2.78
A ♠ lead appears best.
Kudos to Endymion77 for suggesting the winners.

I did a simulation myself

2 runs, one for MP leads and one for IMP leads, each with 1000 deals.

Specification

East: (15-17) balanced, 5 card major possible
West: (9-15), not necessarily balanced, at most 3 cards in each major, except when 4M333.

Result:

IMPs: Contract was beatable double dummy on 258 deals (26%)

Number of deals in brackets how often a specific South card as opening lead would beat the contract:

♠T (123) ♠8 (128) ♠3 (125) ♥Q (80) ♥7 (123) ♥5 (124) ♥2(124) ♦K (108) ♦J (101) ♦8 (142) ♦4 (148) ♦3 (148) ♣4 (108)

MP:

Number of times a specific South card as opening lead would provide most tricks to the defense:

♠T (659) ♠8 (669) ♠3 (667) ♥Q (493) ♥7 (659) ♥5 (662) ♥2 (662) ♦K (394) ♦J (407) ♦8 (487) ♦4 (498) ♦3 (498) ♣4 (632)

My Conclusion:

There is not much to choose between a spade and a heart lead.
A diamond lead is terrible at Matchpoints. In fact it is worse than any other lead except the ♥Q

At IMPs it is close.
While a low diamond lead beats the contract more often than any other lead, the difference compared to a major suit lead is not earth shattering, 12.4% versus 14.8%, a 2.4% difference.
However, you often will lose a trick when a diamond will not beat the contract.
The 2.4% of times you beat the contract more often, and win 11 IMPs has to be balanced against the 16.5% of times where you will lose one or more IMPs in tricks.

Rainer Herrmann

[Zelandakh]

 rhm, on 2014-June-03, 04:28, said:

IMPs: Contract was beatable double dummy on 258 deals (26%)
♠T (123) ♠8 (128) ♠3 (125)

MP:
♠T (659) ♠8 (669) ♠3 (667)

Anyone else find it interesting that the best spade in either case is the 8 from T83? Is that a mini-win for 2/4 or Busso over 3/low? Anyone got a list of holdings where the traditional lead is often wrong?

[billw55]

 Zelandakh, on 2014-June-03, 05:26, said:

Anyone else find it interesting that the best spade in either case is the 8 from T83? Is that a mini-win for 2/4 or Busso over 3/low? Anyone got a list of holdings where the traditional lead is often wrong?

I think difference is small enough to consider that a coincidence.

I am however, astonished by the difference in total tricks between a spade and a diamond.

[rhm]

 billw55, on 2014-June-03, 06:19, said:

I think difference is small enough to consider that a coincidence.

I am however, astonished by the difference in total tricks between a spade and a diamond.

Just to clarify:
The numbers are not total tricks, but how often out of 1000 deals this lead will give the defense the maximum number of tricks on a double dummy basis.
For example the ♠8 will allow the defense double dummy the maximum number of defensive tricks in 669 deals of 1000, or roughly in two thirds of the deals, while the ♦4 or 3 will give you the maximum number of defensive tricks only on 498 deals of 1000 or roughly on one half of all the deals.

Rainer Herrmann

[beatrix45]

The fourth best ♦ is going to be led at the other table(s) most likely. You beat 3NT off the top if partner has the ♦ ace and RHO has Qxx. The other leads are not terrible, but unless you are up against better players and really need a board, they are considerably inferior, imo.

Simulations are very difficult and extremely labour intensive to do correctly. I am utterly unconvinced by anecdotal so-called 'evidence' from such 'research'.

[BillPatch]

 rhm, on 2014-June-03, 04:28, said:

. . . .
At IMPs it is close.
While a low diamond lead beats the contract more often than any other lead, the difference compared to a major suit lead is not earth shattering, 12.4% versus 14.8%, a 2.4% difference.
However, you often will lose a trick when a diamond will not beat the contract.
The 2.4% of times you beat the contract more often, and win 11 IMPs has to be balanced against the 16.5% of times where you will lose one or more IMPs in tricks.

While I agree that the finish between low spade and low heart is inconclusive, it is customary to compare the diamond lead to the better major suit. It is debatable whether to lead second best from T8x or low(Bird/Anthias recommend low from this combination because they believe that following the conventional lead may aid partner more than the statistical gain from starting unblocking on the first round with the lead of the 8. The low spade beats contract 12.5% and produces the optimal defensive trick count 66.9%. Plugging the low spade numbers into the last sentence by Rainer: "The 2.3% you beat the contract more often has to be balanced against the 19.4% you . . . lose one or more IMPs in overtricks." 17.1 low spade better in tricks -(-2.3) low diamond better at set = 19.4% extra overtricks allowed by diamond lead.

[BillPatch]

 beatrix45, on 2014-June-06, 02:06, said:

The fourth best ♦ is going to be led at the other table(s) most likely. You beat 3NT off the top if partner has the ♦ ace and RHO has Qxx. The other leads are not terrible, but unless you are up against better players and really need a board, they are considerably inferior, imo.

Simulations are very difficult and extremely labour intensive to do correctly. I am utterly unconvinced by anecdotal so-called 'evidence' from such 'research'.

According to Richard Pavlicek's[spelling corrected]website, the expected value from playing tens of thousands of bridge hands at three notrump by experts on vue graph championships after the opening lead is equal to the double dummy analysis value. Bird/Anthias further tens of thousands of hands from club play and several thousand hands simulated single dummy on WBridge5 with similar results.

Therefore we may conclude that given adequate statistical power from our sample we may draw the inference that the expected percentage that a low spade will outscore a low diamond is 17.1%. The extra overtricks saved outweigh the reduced rate of setting the contract by eight to one. Since the preponderance of the evidence indicates that equal strength contestants will win .171 board on average by leading over a field lead of the low diamond. The diamond lead is a pure bridge error.

[beatrix45]

 BillPatch, on 2014-June-08, 16:02, said:

According to Richard Pavlichek's website, the expected value from playing tens of thousands of bridge hands at three notrump by experts on vue graph championships after the opening lead is equal to the double dummy analysis value. Bird/Anthias further tens of thousands of hands from club play and several thousand hands simulated single dummy on WBridge5 with similar results.

Therefore we may conclude that given adequate statistical power from our sample we may draw the inference that the expected percentage that a low spade will outscore a low diamond is 17.1%. The extra overtricks saved outweigh the reduced rate of setting the contract by eight to one. Since the preponderance of the evidence indicates that equal strength contestants will win .171 board on average by leading over a field lead of the low diamond. The diamond lead is a pure bridge error.

Your post might be more convincing if you were able to spell Richard Pavlicek's name correctly. Have you ever actually used his hand generator? To do so correctly you have to cast a wide net that includes all possible hands consistent with the bidding. Then you get to winnow out by hand those that are not, in fact, consistent. In this case with this auction it is very labour intensive. For example, responder with 4-3-3-3 shape, decent spots, and looking at a combined 28-31 HCP won't Stayman. That done, then you have to analyze the remaining hands to see what might happen. Sometimes, declarer will have a choice of lines of play. I can see that you love your computer. Now, you need to learn some statistics and the game of Bridge.

[whereagles]

Pavlicek's hand generator had some statistical bias problems. Better check if the issue has been solved before using it.

[Mbodell]

 BillPatch, on 2014-June-08, 16:02, said:

According to Richard Pavlichek's website, the expected value from playing tens of thousands of bridge hands at three notrump by experts on vue graph championships after the opening lead is equal to the double dummy analysis value. Bird/Anthias further tens of thousands of hands from club play and several thousand hands simulated single dummy on WBridge5 with similar results.

Therefore we may conclude that given adequate statistical power from our sample we may draw the inference that the expected percentage that a low spade will outscore a low diamond is 17.1%. The extra overtricks saved outweigh the reduced rate of setting the contract by eight to one. Since the preponderance of the evidence indicates that equal strength contestants will win .171 board on average by leading over a field lead of the low diamond. The diamond lead is a pure bridge error.

Even if over the full population double dummy on average matches single dummy post opening lead, that doesn't mean it will work for any particular hand. It could be there are offsetting biases and that it does poorly with long suits headed by-KJ underestimating the cost of leading from them, but has some other thing that counteracts that over the whole population of hands.

[Mbodell]

 BillPatch, on 2014-June-08, 14:46, said:

While I agree that the finish between low spade and low heart is inconclusive, it is customary to compare the diamond lead to the better major suit. It is debatable whether to lead second best from T8x or low(Bird/Anthias recommend low from this combination because they believe that following the conventional lead may aid partner more than the statistical gain from starting unblocking on the first round with the lead of the 8. The low spade beats contract 12.5% and produces the optimal defensive trick count 66.9%. Plugging the low spade numbers into the last sentence by Rainer: "The 2.3% you beat the contract more often has to be balanced against the 19.4% you . . . lose one or more IMPs in overtricks." 17.1 low spade better in tricks -(-2.3) low diamond better at set = 19.4% extra overtricks allowed by diamond lead.

You have to be a little careful with the IMPs cost of an overtrick or undertrick. It is tempting to just compare to other people in the same contract who make one of the two choices (diamond lead against 3nt or spade lead against 3nt), but it could be the overtricks and undertricks are worth nothing if the other table is in slam down or in a partscore. Just one reason why playing to make/set is important.

[rhm]

 beatrix45, on 2014-June-06, 02:06, said:

The fourth best ♦ is going to be led at the other table(s) most likely. You beat 3NT off the top if partner has the ♦ ace and RHO has Qxx. The other leads are not terrible, but unless you are up against better players and really need a board, they are considerably inferior, imo.

Simulations are very difficult and extremely labour intensive to do correctly. I am utterly unconvinced by anecdotal so-called 'evidence' from such 'research'.

I have done lots of simulations and I beg to differ.
When I publish my simulation results I always specify my specifications, but I also know from experience there will always be some, who will critique my specifications or double dummy simulation results on general principles.
As a rule I keep my specifications simple and do not worry about the odd hand.
What I have learned from experience and experimentation is that worrying about rare exceptions are not worth it, because the final result is not affected in a significant way.
As a rule dealing out more hands does more to the precision of the result.
When you do lots of simulations you get a feeling what might affect and and what will not affect the outcome.
Once you have this experience you can do simulations properly without them being labour intensive. Good Bridge judgement, a basic understanding about statistics and logic suffice.
Of course there are bidding scenarios, which can not be simulated easily, usually because people differ too much about the requirements for a bid.
The above is not one of them.

I am also pretty sure that advances in bidding at the top level in the last years is in no small part due to research done based on double dummy simulations.
Those, who are over critical to simulations only handicap themself.

Rainer Herrmann

[whereagles]

Well, I regularly simulate stuff when I design my bidding systems.

So far so good w.r.t. results

[PhantomSac]

 Mbodell, on 2014-May-15, 03:39, said:

I choose a ♠ at IMPs and a ♥ at MP in a good field, ♦ at MP in a poor field.

At IMPs I'm trying to beat the contract and am assuming they barely have game values so partner has more of our values. I also assume that we are likely to be longer in the majors thanks to the auction (not a guarentee, but more likely than not, especially if they are around 25 or 26 hcp as opposed to 28 or 29 that would be even more likely to suppress the 4M in 4333). So therefore I lead the ♠.

In a weak field at MP I don't want to lose the board on the opening lead and "everyone" will be leading 4th best diamond.

In a strong field I'm not in love with leading from KJxxx in a minor, so I have to choose between the heart and the spade. It is no longer the case that I'm playing just to set. Holding them to 9 when many make 10 or 10 when many make 11 might be good enough. The risk in leading the heart is less than the diamond IMO. Maybe I should lead the spade too as there is risk from leading the heart, but we may need the tempo to get a heart trick before they run clubs and top spades, so I'll try that. Of the three situations this is the one I'm least confident is right.

A spade at imps? Do you guys actually lead like this? What kind of convoluted thinking makes you believe that a SPADE is the most likely lead to set 3NT? It's just absolutely absurd I'm sorry.

[PhantomSac]

As usual with these double dummy simulations, do any of you discount the hands that partner will double 3N with? Definitely some of the hands that a spade binks partner will X 3N or possibly have doubled 3N. Likewise, Rainer has the 3N bidder as "not necessarily balanced." That's great but I do not remember the last time a good player bid 1N 3N with a small singleton in a major, let alone with a void. Maybe in the old days, but nowadays most people play something at least like transferring to a minor and bidding 3M as short, or 1N 3M short, etc. Those hands should be excluded. Finally, in the real world partner will often fail to find a killing diamond shift if he gets in. I realize double dummy analysis cannot do anything about this but there is something to be said for leading your good 5 card suit making the defense easier.

But really not excluding partner having a double of 3N, and not excluding 1N 3N bids that have a stiff major are pretty inexcusable when doing analysis like this, and even failing to do that shows that the diamond lead will more likely beat it.

Forget about real world considerations like sometimes accidents happen if you lead the spade 8 and can run diamonds later (eg partner having A A in the pointed, not likely but just to illustrate, good luck getting the D shift 100 % of the time on your spade lead), or sometimes you will set up diamonds and they will need to guess the hearts which they always will double dummy, or they will need to pick up your partners club holding with a deep finesse, etc (hint: setting up a long suit is a big threat that is less big in double dummy analysis), or the fact that a spade lead will sometimes blow the suit in real life but seems passive to a double dummy analysis, a diamond is STILL more likely to beat it according to this analysis!

[PhantomSac]

 rhm, on 2014-June-03, 04:28, said:

At IMPs it is close.
While a low diamond lead beats the contract more often than any other lead, the difference compared to a major suit lead is not earth shattering, 12.4% versus 14.8%, a 2.4% difference.
However, you often will lose a trick when a diamond will not beat the contract.
The 2.4% of times you beat the contract more often, and win 11 IMPs has to be balanced against the 16.5% of times where you will lose one or more IMPs in tricks.

Rainer Herrmann

If your lead blows a trick 16.5% of the time double dummy that does not mean you lose an overtrick imp 16.5 % of the time. A funny thing happens at imps when you lead your long/good suit, once it is set up people often start cashing out. This is one flaw in double dummy analysis. But even with your very flawed simulation that is completely biased towards the passive lead (much like Bird), gaining 11 imps 2.4 % of the time is worth .264 imps, compared to gaining 1 imp 16.5 % of the time which is worth .165 imps, so you gain .1 imps a board by leading a diamond vs a spade. This is completely convincing to me that a diamond is much better given your simulation parameters.

[beatrix45]

I have given some thought as to how long a proper analysis of this problem might take assuming I were already set up to do this particular problem. It looks like 8-10 hours minimum, if I were very lucky.

1. (1hr) Generating 200-500 hands using the Pavlicek generator. Example: RHO has 15-17 HCP, no singleton and no 6+ card major suit. LHO has 10-15 HCP, no 5 card major and no singleton or void. Partner gets what is left.

2. (2hr) Remove any hands not fitting the actual bidding. A few of these will be removed because RHO would not have opened 1NT, but most will be hands where LHO might have used Stayman or at least not have passed. In a few cases partner might have a double. This should only take a few seconds per hand in most cases, but we have a lot of hands to examine.

3. (4hr) Analyze the remaining hands for their play. I would expect to have 50 to 100 hands left to analyze. Most will take only a minute or so, but a few will be difficult. This is particularly true when declarer or partner or your hand has a choice to make in the play. You cannot assume double dummy play. In the end, for a few hands you may decide to assign probabilities to the various possible outcomes. You probably should go back and use a computer playing program to vet your results.

4. (2hr) Translate your results into IMP or MP scores and summarize.

Simulations can be a wonderful tool, but if you don't do them correctly, they are worse than useless. IF YOU INTRODUCE SAMPLE BIAS OR FAIL TO ANALYZE CORRECTLY, YOU GET THE WRONG ANSWER, FOOL.

[StevenG]

 beatrix45, on 2014-June-11, 05:13, said:

4. (2hr) Translate your results into IMP or MP scores and summarize.

How do you translate your result into an MP score? Isn't your score almost entirely dependent on what happens at other tables?