[helene_t]

I am not saying that DD analysis is irelevant. Surely, absent other evidence, one should take the advice from the DD results when chosing opening lead strategies.

As for how strongly DD research should be weighted against possible contrary evidence, I don't really have a qualified opinion about it.

All I am saying: I simply don't see any relevance at all to the fact that average DD results are similar to average SD results for game-level contracts. Would it mean that, given that declarer has an advantage SD over DD in low-level contracts while the opposite is true for slams, the DD research would be less relevant for choosing opening leads against slams and partscores? I don't see why, as far as I am concerned the DD research, as far as opening leads is concerned, would be equally relevant regardless of how good DD research was at predicting other things.

If (I do say "if") it is so that passive leads work better DD than SD, I can see why that might be so: a passive lead can help the SD declarer locate a queen in partner's hand but it won't help the DD declarer. Compare that to a lead from KJxx. If it costs a trick SD it probably also costs a trick DD. Also, a passive lead works better when partner always can recognize it as such, and even find the right switch.

OTOH, that the lead of an unsupported ace works so well DD is intriguing. The argument that it works well because DD it is always followed by the right switch, doesn't sound credible to me. Often the right switch will be found SD as well. The strategy to lead an ace in order to get a look at dummy (and at partner's signal) is something that makes sense SD but not DD so if anything I would expect unsupported aces to work even better SD than DD.

I was just looking at a hand from this month's EBU magazine. You are in 3NT with AJ9 of hearts in dummy and 8 sec in your hand. You receive a small heart lead. You play opening leader for the 10 so you play the 9. If he made a standard lead from KQxx(x) you are down. DD you would have made it regardless of lead. And if he had made the (DD-research-favored) lead of the king you would have made it SD as well.

[mike777]

I am not saying that DD analysis is irelevant. Surely, absent other evidence, one should take the advice from the DD results when chosing opening lead strategies.


ok and?

ok I put you on the stop...you don't know....

perhaps I don't understand ...you answer is you don't know?

May I assume I never or almost never have said evidence?

In other words should I always take the advice of DD to win?

[helene_t]

mike777, on 2013-October-14, 00:14, said:

In other words should I always take the advice of DD to win?

What about agreeing with p to play top-of-nothing against both suit and notrump so that p can recognise a passive lead? And K from suits headed by AK so that he can recognise an unsupported ace? And agree with p that you lead an honour from HHxx(x) so that he doesn't necessarily expect three honours? Give it a try and see if you like the results. And share your experience with us.

[dake50]

Bounce around as much as you like.
DD simulators SHOULD tabulate opening leads
-- agreeing Goren, Journalist, Rusinow, Combine, etc OR disagreeing each.
Then tabulate those scores.
Who has done that??
It's been computationally available for decades -- just not done.!

[PhilKing]

dake50, on 2014-July-01, 09:21, said:

Bounce around as much as you like.
DD simulators SHOULD tabulate opening leads
-- agreeing Goren, Journalist, Rusinow, Combine, etc OR disagreeing each.
Then tabulate those scores.
Who has done that??
It's been computationally available for decades -- just not done.!

I might be being a bit thick, but in a DD simulation I think the result would be an exact tie.

[awm]

Having now read these books, I think they are in fact very good.

The authors have several ways to deal with the objections to double dummy:

1. They run sims to determine the frequency of the common "double dummy" positions people complain about where the lead solves a guess for declarer. All indications are that these positions are pretty infrequent and do not greatly change the results.

2. They attempt to infer logical reasons for the leads that succeed. For example, they observe that if the auction goes 1NT-3NT and I have a five-card minor, partner is very likely to be extremely short in that minor. This justifies the poor showing of the "lead from length" in this situation (leading from two or three small in a major is often better).

3. They compute the chance of partner having specific cards, opposite my lead from a given hand and a given auction. When opponents have most of the strength and most of the outstanding cards in a particular suit, your chance of hitting partner with the "magic" holding is really quite poor.

In general I find the objections to double dummy simulation unconvincing. The sims have a lot of data on their side. The objections have only anecdotal evidence, with no serious attempts at producing data.

[whereagles]

Thing is, gathering data for real-life dummy play isn't as simple as turning the crank. It requires a lot of time and research to get examples from expert play. But yeah I agree that DD objectors need it to be taken seriously.

Discrediting science is routine in the first times but in the long run proving its simplifying assumptions are insufficient requires evidence. Blind faith also works, but you need a continuous supply of gurus to keep people "under control".

[sfi]

So my partner and I separately each read one of these books a while ago and the results, while anecdotal, are illuminating. All play was at national competitions.

I read the leading vs. suit contracts and it clearly improved my leads. There were several examples where I beat contracts I would have not otherwise done so, and only one instance I can remember where my previous choice would have worked out better. My partner didn't notice any change in leading style, so it did not affect his decision-making process.

He read the leading vs. NT contracts a few months earlier. The very next tournament I started noticing that my inferences from the opening lead were often not valid, which meant that my defence was frequently less than optimal. I don't know whether his lead choices were an overall net gain or loss, largely because I did not realise why they had changed until the middle of the second tournament.

I finally started reading the leading vs. NT contracts book before the last tournament and it all made much more sense. At last I could see why he was making the choices and that made the defence more comfortable. And I even paid attention when I picked up the following hand to lead against a 1NT - all pass auction:

A963
AQ87
954
Q5

The passive diamond lead hit partner's long suit and we beat the contract 3 tricks on a hand where our teammates made an overtrick in the same contract on a different lead (it doesn't really matter which one).

So, what conclusions have I drawn?
1. Defence against suits is much less about cashing length tricks, so it is much less important to hit the side's best suit. This means that the types of inferences drawn from the opening lead vary greatly between the two books, and it's easier to apply the leads vs. suits without involving partner.
2. There may be a double dummy bias in the analysis, but the practical advice has achieved positive outcomes in most cases.
3. Not letting partner know about your change in leading tendencies can throw the defence off. Talk to partner about the theory, particularly if you are going to stop leading 4th best from your long suits.

My partner thinks the NT book gives more valuable advice, while I think the suit book does. I wonder if this is because we each prefer the one we read first or whether there is a substantive difference between them.

[PrecisionL]

Yes, the NT book confirmed the style of opening leads my partnership has been using for 3 years!

[whereagles]

Excellent post there, sfi.

The books results clearly hint at a change in lead style and, consequently, at lead conventions.

At NT, the old "low from HHxx" needs to be changed to "top card from HHxx".
At suit contracts, since leading unsupported aces is important, you might want to try Rusinow leads.

[Vampyr]

sfi, on 2014-July-02, 05:35, said:

3. Not letting partner know about your change in leading tendencies can throw the defence off. Talk to partner about the theory, particularly if you are going to stop leading 4th best from your long suits.

You seem to give short shrift to disclosure to the opponents.

[sfi]

Vampyr, on 2014-July-02, 08:59, said:

You seem to give short shrift to disclosure to the opponents.

Not sure how you came to that conclusion. When was the last time you asked an opponent how likely they are to lead a short major against NT vs. a 5 card minor? What sort of an answer would you expect?

In any case, you can't have an agreement to disclose if you don't first talk about it with your partner.

[barmar]

Reading these books will probably also help your defense when playing in BBO robot tourneys. Their defensive strategy is based on DD simulations similar to that employed by the books (although far fewer hands are dealt, due to time constraints). Whenever someone complains "Why don't the robots return my suit when I lead 4th best against NT?", I refer them to the book, which suggests that these leads are not necessarily best. It tends to avoid leading away from honors, preferring to make a passive lead and hoping that partner will get in and find the killing switch.

[barmar]

Vampyr, on 2014-July-02, 08:59, said:

You seem to give short shrift to disclosure to the opponents.

I don't think you have to disclose how you decide which suit to lead, that's just bridge judgement.

You have to disclose agreements about styles of leads, e.g. Rusinow versus top of an honor sequence, and 4th best versus 3rd/5th.

Note also that an agreement to lead 4th best doesn't mean that the lead of a low card promises at least 4 cards in the suit, or that you must lead from a 4-card suit. It just means that if you do have at least 4 in the suit you lead, you'll lead the 4th highest (unless you have a sequence at the top and you choose to make a sequence lead instead).

[awm]

Here are some possible data points which could refute this book (if they held):

1. Examine some set of hands from real play where the same contract was defended and different leads were made. Or do the same with "single dummy" simulation. Show that some leads perform consistently better/worse single dummy than they would double dummy.

2. Enumerate a set of situations where leads which do not cost double-dummy in fact help declarer at single dummy. Use a hand generator to show that these situations actually arise at a high level of frequency.

3. Analyze hands from top-level play to find examples where the defense dropped a trick subsequent to the opening lead. Show that this is quite common, and that it's more frequent on passive leads.

I haven't seen really serious attempts at any of the above. In fact I've seen analysis which suggests that 2 is not true (from the Bird/Anthias books in fact) and that 3 is not true (numerous measures of "declarer advantage" over double dummy both before and after the opening lead).

There are some minor quibbles with the book, in that the bidding of the authors differs from what I'd expect in some common auctions. For example they suggest that after 1M-1NT (where 1NT is non-forcing or semi-forcing) that opener will pass with most minimum 5M-4m hands, whereas in my experience most players will bid on these hands. They also assume most players will stayman with (43)33 type hands whereas my experience is the opposite (although they do give evidence that this does not make a huge amount of difference). Of course some partnership style in the auctions is unavoidable.

[jallerton]

barmar, on 2014-July-02, 10:01, said:

I don't think you have to disclose how you decide which suit to lead, that's just bridge judgement.

Quote

LAW 40 - PARTNERSHIP UNDERSTANDINGS

A. Players’ Systemic Agreements

1. (a) Partnership understandings as to the methods adopted by a partnership may be reached explicitly in discussion or implicitly through mutual experience or awareness of the players.

(b) Each partnership has a duty to make available its partnership understandings to opponents before commencing play against them. The Regulating Authority specifies the manner in which this shall be done.

Quote

LAW 20F2. After the final pass and throughout the play period, either defender at his own turn to play may request an explanation of the opposing auction. At his turn to play from his hand or from dummy declarer may request an explanation of a defender’s call or card play understandings. Explanations should be given on a like basis to 1 and by the partner of the player whose action is explained.

If, like sfi, you have discussed your leading style with partner than you have an explicit partnership understanding. If you have played with someone for a while and notice what they tend to lead in certain situations then you have an implicit partnership understanding. Laws 40A1b and Law 20F2 imply that these understandings should be disclosed to the opponents.

If you know that your partner has an aggressive style of overcalls, you would disclose that fact to the opponents, would you not? If you know that your partner has an aggressive style of opening leads, why is that any different?

[barmar]

jallerton, on 2014-July-02, 16:17, said:

If, like sfi, you have discussed your leading style with partner than you have an explicit partnership understanding. If you have played with someone for a while and notice what they tend to lead in certain situations then you have an implicit partnership understanding. Laws 40A1b and Law 20F2 imply that these understandings should be disclosed to the opponents.

If you know that your partner has an aggressive style of overcalls, you would disclose that fact to the opponents, would you not? If you know that your partner has an aggressive style of opening leads, why is that any different?

So what kind of disclosure would you expect? "He's less likely to have led away from an honor than most other players"? That's probably about as specific as you can get about something like this.

[Cascade]

Whatever information you have. The example you give should be routine but is seldom disclosed.

I think it is bizarre that some people think they can play without disclosing their partnership understandings including implicit ones and partnership experience.

[FM75]

awm, on 2014-July-01, 21:44, said:

Having now read these books, I think they are in fact very good.

The authors have several ways to deal with the objections to double dummy:

1. They run sims to determine the frequency of the common "double dummy" positions people complain about where the lead solves a guess for declarer. All indications are that these positions are pretty infrequent and do not greatly change the results.

2. They attempt to infer logical reasons for the leads that succeed. For example, they observe that if the auction goes 1NT-3NT and I have a five-card minor, partner is very likely to be extremely short in that minor. This justifies the poor showing of the "lead from length" in this situation (leading from two or three small in a major is often better).

3. They compute the chance of partner having specific cards, opposite my lead from a given hand and a given auction. When opponents have most of the strength and most of the outstanding cards in a particular suit, your chance of hitting partner with the "magic" holding is really quite poor.

In general I find the objections to double dummy simulation unconvincing. The sims have a lot of data on their side. The objections have only anecdotal evidence, with no serious attempts at producing data.

For the most part, I agree. But a data scientist would strongly object to their publication.
1) They do not supply their data for independent review.
2) Their experiments are not "repeatable" (exactly) - can't reproduce the exact deals that they used.
3) Their experiments are not "repeatable" (even approximately), because they do not specify the constraints imposed on the deals.
4) They do not compare their results properly against a "null hypothesis". Was it better than the alternative - with respect to both correct predictions as well as false positives and negatives? By how much?
I laud the approach. I have criticized elsewhere the poor proof-reading - and obviously the errors missed in the second publication.

Kudo's for a scientific approach to the problems.

[Mbodell]

FM75, on 2014-July-03, 16:46, said:

For the most part, I agree. But a data scientist would strongly object to their publication.
1) They do not supply their data for independent review.
2) Their experiments are not "repeatable" (exactly) - can't reproduce the exact deals that they used.
3) Their experiments are not "repeatable" (even approximately), because they do not specify the constraints imposed on the deals.
4) They do not compare their results properly against a "null hypothesis". Was it better than the alternative - with respect to both correct predictions as well as false positives and negatives? By how much?
I laud the approach. I have criticized elsewhere the poor proof-reading - and obviously the errors missed in the second publication.

Kudo's for a scientific approach to the problems.

How often does anyone manage to do all of that? I strongly agree that, at the very least, sharing code and underlying assumptions about auctions and holdings is very important. I think we (the computer literate/simulation literate bridge players) could very easily do loose peer review of trying to recreate some of their results to spot check do we get the same results as them, and if not, why not. There might be value there both in confirming their results and also seeing how differing assumptions (like no stayman with 4333) might change (or not change) the overall results.