Bird/Anthias books on opening leads

[whereagles]

Hi guys,

 

It's been a while since I showed up in the foruns (got sunk into a research project and didn't yet come back to surface... lol). Anyway, I was wondering whether people have been reading David Bird/Taf Anthias books on opening leads. If so, are you changing methods because of it? The books are:

 

Winning NT leads

Winning trump leads

 

(For those unfamiliar with the books, the authors simulate outcomes of various opening leads with a double-dummy analyser.)

 

Bottom line is that the authors find some outcomes that contradict current theory. Here are some results:

 

1 - Leading an unsupported ace works much better than textbooks suggest. These leads very often rank as #1.

2 - In NT leading an honor from HHxx+ (touching honors) works better than the usual 4th best lead.

3 - Doubleton and singleton leads also work better than textbooks suggest. The danger of helping declarer setting up a side suit is overrated.

4 - Aggressive leads from Kxx or Qxx are very likely to blow a trick.

5 - Leads from Jxx are almost as safe as leading from xxx and should be preferred to Hxx leads.

 

My question regards issues 1 and 2. Are people switching to lead methods like e.g. Rusinow to be able to lead unsupported aces more often? Are people redesigning their NT honor leads?

 

Thx in advance for any comments.

whereagles

[PrecisionL]

Yes, I have both books (one print and one e-book) and basically agree with their analysis.

 

Yes, I have been playing such methods several years based on my own experience and analysis (thanks to hand records).

 

I have tried to teach a few of my partners, it is hard to get them to change.

 

1. No, I don't lead unsupported Aces (unless gambling NT).

 

2. I lead 2nd (Hxxx) and 4th (QTxx or better) unless 3-cd Major seems better.

[jeffford76]

Much of the gain in ace leads in double dummy simulations is that you always switch to the correct thing when continuing the suit is wrong. That said, I've had a fair amount of success leading them more often against NT. In the ACBL the standard method is to lead the ace to ask for an unblock, so it's unsurprising that aces aren't led often as doing so requires changing methods. I've been happy with the switch to Rusinow.

[benlessard]

I believe that double dummy simulation are generally useful however leads is where I have the most doubt.

 

But im pretty sure I will still enjoy the book.

[helene_t]

Much of the gain in ace leads in double dummy simulations is that you always switch to the correct thing when continuing the suit is wrong. T

Yes. Also, the advantage of the passive lead is that partner always find the right switch. And leading an honour from HHxx works better when partner can read when you have KQ32 and when you have KQT2.

 

They might very well be right but it would take some SD analysis to convince me.

[MickyB]

Yes. Also, the advantage of the passive lead is that partner always find the right switch. And leading an honour from HHxx works better when partner can read when you have KQ32 and when you have KQT2.

 

They might very well be right but it would take some SD analysis to convince me.

 

And leading passive works better when you can't solve a guess for declarer

 

Leading low from KQxx works less well when declarer can't misguess and play you for KTxx/QTxx

 

I asked David if his book relied solely on DD analysis, and he said yes, but that was justified in the book. I haven't seen the justification but I can't imagine I would find it persuasive.

 

The bridge world seems to be divided into those that never lead from kings and those that never lead from jacks. Apparently the book strongly supports the latter group, can anyone think of reasons why DD analysis would favour them?

[rhm]

And leading passive works better when you can't solve a guess for declarer

 

Leading low from KQxx works less well when declarer can't misguess and play you for KTxx/QTxx

 

I asked David if his book relied solely on DD analysis, and he said yes, but that was justified in the book. I haven't seen the justification but I can't imagine I would find it persuasive.

 

The bridge world seems to be divided into those that never lead from kings and those that never lead from jacks. Apparently the book strongly supports the latter group, can anyone think of reasons why DD analysis would favour them?

Bird does give justification and looks at objections.

As far as I can tell the objections to DD leads are essentially the same as to DD simulations in general.

Most are valid on specific layouts like declarer never going wrong on a two way guess etc.

But even if, how many of the total deals are really affected, how much does it affect the overall results?

The answer seems to be, marginally at best. The defense has as many guesses it will get right double dummy.

 

Another objection and its refutation:

 

Of course if you lead an ace and look at dummy you might be not much wiser.

But in the vast majority of cases, looking at trick one and the remaining cards in dummy, should make your subsequent choice on average better, even if you will not always find the deadly switch.

Of course there is a cost. There may not be any deadly switch left, otherwise people would always lead an ace.

And what seems to be always poorly understood, yes DD leads assume you will find thereafter the optimal continuation.

However, this is balanced by the fact, that declarer will also not make any errors in getting his tricks, hardly the case in practice.

What critics rarely grasp is that these opposite effects seem to cancel out each other.

 

I am not a blind believer in DD simulations, but I do think they are useful.

The critics, Michael Rosenberg for example, who is a great player and who I respect, leave me unperturbed.

 

Rainer Herrmann

[gwnn]

What critics rarely grasp is that these opposite effects seem to cancel out each other.

Could it be that critics rarely grasp this because proponents do not support this statement with any substantial evidence?

[blackshoe]

Much of the gain in ace leads in double dummy simulations is that you always switch to the correct thing when continuing the suit is wrong. That said, I've had a fair amount of success leading them more often against NT. In the ACBL the standard method is to lead the ace to ask for an unblock, so it's unsurprising that aces aren't led often as doing so requires changing methods. I've been happy with the switch to Rusinow.

It surprised me to learn that many modern experts lead Rusinow against NT, considering the (pretty good, I thought) argument against doing so made by "The Journalist" about half a century ago. I'm still not sure why the moderns think it's an improvement.

[rhm]

Could it be that critics rarely grasp this because proponents do not support this statement with any substantial evidence?

This is just not true, but critics just do not want to listen.

 

For you I repeat one more time what the evidence is:

 

The evidence the proponents put forward is, that over many boards played single dummy by humans the trick taking result achieved by double dummy is remarkable close to the one accomplished by single dummy play at the table.

Any mathematician or scientist with a basic knowledge in statistics will tell you that this is a sound basis for doing simulations with many deals double dummy to find out what will be the likely outcome single dummy.

If we look where the differences are largest we find:

Declarer has a slight advantage in low level notrump contracts compared to DD results.

The defense has a slight advantage in slam contracts.

Even there the difference is not earth shattering.

 

Nobody claims that the way double dummy gets its tricks is the same as single dummy. But this is irrelevant for the result. Yet this is the line by which critics attack DD simulations time and again.

 

Rainer Herrmann

[gwnn]

Could you link me to one of these studies? I have heard this assertion several times but I would like to see the actual numbers. Maybe I can even find some scientists here in the University where I work (and post from...) and ask them about vague references to data without citing it.

 

I would guess that no one did a study of unsupported ace leads in SD vs DD setting for example. Last time you linked to a DD study it apparently proved that 4333's are stronger hands than "5322" hands. I trust that you realise that it was not exactly a peer-reviewed journal-worthy research.

[helene_t]

Pavlicek has some statistics of DD compared to SD but I am not sure how relevant it is. Even if the average tricks taken in notrump contracts are similar SD to DD, it doesn't follow that the SD=DD equation holds for the statistic we are discussing, namely

(tricks taken when a passive lead is chosen) - (tricks taken when an aggresive lead is chosen).

 

If you watch GIB's play you will find some that are obviously inferior SD but could (and presumably are, otherwise GIB wouldn't chose them) OK in a DD analysis. For example, GIB often leads a stiff king of trumps. So claiming that a DD analysis is adequate for every lead strategy would be absurd.

 

The question is if a DD analysis is adequate for particular lead styles, such as unsupported aces, broken suits, HHxx, small tripletons. I can't prove that there is no evidence for this but if there really was robust SD data available, wouldn't Bird/Anthias have used them?

[rhm]

Could you link me to one of these studies? I have heard this assertion several times but I would like to see the actual numbers. Maybe I can even find some scientists here in the University where I work (and post from...) and ask them about vague references to data without citing it.

 

I would guess that no one did a study of unsupported ace leads in SD vs DD setting for example. Last time you linked to a DD study it apparently proved that 4333's are stronger hands than "5322" hands. I trust that you realise that it was not exactly a peer-reviewed journal-worthy research.

My conclusion from the DD study I cited was that in notrump contracts distribution matters little and the difference between 5332 and 4333 in notrump below the slam level is much less than common wisdom assumes. Many DD studies I have seen agree.

Of course if you find out in the bidding that you have a good fit with a long suit and all other suits to be well stopped you should upgrade.

But in the bread and butter case, where you consider raising a 1NT opening to 3NT, I do prefer to have 5332 to 4333, but in the knowledge that the difference is much less than what the world perceives it to be.

I will tend to be less optimistic with 5332 but less pessimistic with 4333 than what is common wisdom, without going to the extreme to assume that 4333 is better than 5332.

 

Few research in Bridge qualify for peer-reviewed jounal-worthy research. This does not prove this research wrong.

 

Some, which did, were Bridge nonsense. e.g. 36. Cowan, R. Hand evaluation in the game of Contract Bridge. J. Royal Statistical Society: Series C; Applied Statistics 36 58-71 (1987).

 

For example:

 

http://crystalwebsite.tripod.com/double_dummy_accurate.htm

 

This one is old and many years available on the web.

 

Rainer Herrmann

[gnasher]

http://crystalwebsit...my_accurate.htm

The only thing I could see here was a single page which compared the average number of tricks double-dummy with the average number of tricks single-dummy, across all hands. That doesn't tell us anything about the merits of using double-dummy analysis to determine the best opening lead.

 

Here's another thought-experiment for you to dismiss. Suppose that with a given hand and auction:

- Every human would lead Card A

- Leading Card A beats the contract 30% of the time single-dummy, but 20% of the time double-dummy.

- Leading Card B beats the contract 20% or the time single-dummy, but 30% of the time double-dummy.

 

This would be consistent with the study that you cite: the double-dummy number of tricks is the same as the single-dummy number of tricks. However, double-dummy analysis would tell us to lead Card B, and it would be wrong.

[rhm]

Pavlicek has some statistics of DD compared to SD but I am not sure how relevant it is. Even if the average tricks takes in notrump contracts are similar SD to DD, it doesn't follow that the SD=DD equation holds for the statistic we are discussing, namely

(tricks taken when a passive lead is chosen) - (tricks taken when an aggresive lead is chosen).

 

For example, GIB often leads a stiff king of trumps. So claiming that a DD analysis is adequate for every lead strategy would be absurd.

 

The question is if a DD analysis is adequate for particular lead styles, such as unsupported aces, broken suits, HHxx, small tripletons. I can't prove that there is no evidence for this but if there really was robust SD data available, wouldn't Bird/Anthias have used them?

 

I do not know how GIB works, but it is clear that it does not play double dummy and if it does simulate double dummy in single dummy play it does this on very small samples and accordingly with a large margin of error.

That is because, GIB plays way too fast for meaningful simulations during the play. Modern computer power is just not capable of doing this in the allotted time.

I am pretty sure it more relies on some artificial inferences, which are still mediocre compared to human experience and human inference capabilities.

So GIB proves nothing.

 

With regards to leads: Whenever a certain aspect of the game is researched with double dummy analysis, the question comes up whether for this particular question DD results are valid.

I have only seen comparisons, where DD results have been compared to single dummy results broken down to contract level and strain.

There it worked amazingly accurate.

I have heard no good reasons or arguments, why the above relationship should suddenly break down and big discrepancies between double dummy and single dummy occur when certain aspects of the game are analysed.

I have seen little data and software for analyzing single dummy results broken down to certain aspects like leads from broken suits etc.

This is not trivial to do.

Bridge Browser was a software in this direction, but I do not have it and in the last years little was heard about it any more.

Deals selected with Bridgebrowser according to criteria of interest from actual play would need to be fed into a double dummy analyzer like deep finesse and the results compared.

 

Rainer Herrmann

[rhm]

The only thing I could see here was a single page which compared the average number of tricks double-dummy with the average number of tricks single-dummy, across all hands. That doesn't tell us anything about the merits of using double-dummy analysis to determine the best opening lead.

 

Here's another thought-experiment for you to dismiss. Suppose that with a given hand and auction:

- Every human would lead Card A

- Leading Card A beats the contract 30% of the time single-dummy, but 20% of the time double-dummy.

- Leading Card B beats the contract 20% or the time single-dummy, but 30% of the time double-dummy.

 

This would be consistent with the study that you cite: the double-dummy number of tricks is the same as the single-dummy number of tricks. However, double-dummy analysis would tell us to lead Card B, and it would be wrong.

A nice thought experiment.

I just wonder whether you yourself believe this thought experiment has much likelihood to align with reality.

I once studied mathematics and I know well that I can not prove in a mathematical sense that this type of concern is unrealistic for a larger number of deals but common sense tells me so.

When I said double dummy derives it tricks differently from single dummy it still follows the basic rules of the game.

 

What I can tell you is that I often analyse interesting deals I played, where errors may not be immediately apparent, with a double dummy analyzer afterwards.

It helps to have an open mind.

Sometimes the double dummy analyzer shows a winning line and sometimes the line is unrealistic.

But more often I start searching for clues single dummy.

It often hurts, but I learn something from it.

 

Rainer Herrmann

[uday]

This is the Pavlicek page , analysing SD vs DD for 66,548 hands

 

 

http://www.rpbridge.net/9x29.htm

[rhm]

This is the Pavlicek page , analysing SD vs DD for 66,548 hands

 

 

http://www.rpbridge.net/9x29.htm

Compared to the the study I quoted, which contains millions of single dummy plays but different to Pavliceks study is not from top level play, the small number of deals left when Pavlicek 66,548 deals are broken down to strain and level the difference of the results between single and double dummy remains remarkably small.

For example there are only 79 deals were the final contract was 7NT, yet the differences is only 0.04 tricks before the opening lead and 0.02 tricks after the lead.

 

Rainer Herrmann

[rhm]

This is the Pavlicek page , analysing SD vs DD for 66,548 hands

 

 

http://www.rpbridge.net/9x29.htm

Compared to the the study I quoted, which contains millions of single dummy plays but different to Pavliceks study is not from top level play, the small number of deals left when Pavlicek 66,548 deals are broken down to strain and level the difference of the results between single and double dummy remains remarkably small.

For example there are only 79 deals were the final contract was 7NT, yet the differences is only 0.04 tricks before the opening lead and 0.02 tricks after the lead.

 

Rainer Herrmann

[benlessard]

If we look at 3nt with DD leads from Pavlicek website

 

its 8946 making for real life while 8410 making for DD.

 

But when human lead +DD continuation = 8619 real life and 9054 human lead+DD.

 

In short.

 

3NT will make slightly less than 70% IRL.

 

3nt will make 65% double dummy all the way

 

3Nt will make 73% with humans on lead and DD afterwards.

 

This either mean that we are pretty poor for leads or that a DD lead isnt real bridge.

[awm]

This either mean that we are pretty poor for leads or that a DD lead isnt real bridge.

 

I don't think either of these conclusions follows. Certainly we are not as good on opening lead as double dummy. However, humans never play as well as double dummy; even a computer with unlimited processing power won't play as well as double dummy because of incomplete information. Arguably the opening lead is the toughest part of the game because we have least information (can't see dummy, no signal from partner yet, etc).

 

The fact that overall human results are close to double dummy (especially at the slam level) is interesting but not really indicative. There are hands where declarer is likely to do worse than double dummy (for example, where the only way to make is some anti-percentage play like dropping a stiff king in an eight-card fit). There are other hands where the defense is likely to do worse than double dummy. All the overall human results suggest is that these hands are close to equally common... not what will happen on any particular hand.

 

With that said, I think there is room for improvement on opening leads, and the findings of this book are certainly interesting. It's well-known though that both ace leads and passive leads fare better double dummy than single dummy. The main reason is two assumptions: that giving declarer a guess he could make himself will never cost a trick (true double dummy, but not single dummy) and that the defense will never err after the opening lead (again true double dummy, but not necessarily single dummy).

[benlessard]

I agree with 99% of your post except that your argument seems to strengten my conclusion not attack it (i was a sarcastic when we said we are all poor on leads)

 

If you lose one fourth ! of tricks on every 3Nt contract so at the end declarer has 8% extras chance of making the contract how come you cannot come to the conclusion that leading DD and real bridge is totally different ? A fourth is really huge IMO, most of the time I blow a trick on lead im still pretty much convinced ive made the correct lead, so overall im pretty sure we don't do that many mistakes on leads.

 

What seems to work in DD isnt dont seems to based on any real bridge principle, its simply that real bridge lead have a big inherent luck factor that cannot be avoided.

 

Im not saying there is any lesson to be taken from DD, just that the numbers will never really mean a lot because on lot of hands its simply a matter of seing dummy vs not or a matter of luck.

 

Of course I should really read the book before continuing.

[helene_t]

I think it would be very difficult to reach any firm conclusion about effectiveness of different lead styles based on empirical research.

 

If the lead of an unsupported ace tends to work well then it could be offset by a decrease in efficacy of leads of supported aces because of the lack of inference that partner has the King. So we would probably have to compare the success of pairs that adopt one lead style with pairs that adopt another lead style. But obviously it could be that players that lead in a particular way just tend to be better defenders.

 

Maybe if we could identify a large number of pairs in which the two partners have very different lead styles and then see if the results are different when A is on lead than when B is on lead. There are still possibly confounding issues. But if several analyses pointed in the same direction we would have a reasonable indication.

[rhm]

The fact that overall human results are close to double dummy (especially at the slam level) is interesting but not really indicative.

Not indicative of what?

 

There are hands where declarer is likely to do worse than double dummy (for example, where the only way to make is some anti-percentage play like dropping a stiff king in an eight-card fit). There are other hands where the defense is likely to do worse than double dummy. All the overall human results suggest is that these hands are close to equally common... not what will happen on any particular hand.

True. I still do not see what that refutes.

I would never claim that a double dummy line on any single deal has to be the right one. Nor an opening lead

 

But if double dummy would claim for a specific hand over a large number of samples card x wins by a large margin over any other card against some contract, you would have to provide very strong arguments why that is a fluke and another card is superior single dummy in the long run.

We know that different leads win on different possible layouts of the remaining 39 cards, so what is surprising that there is an 8% difference in success between double dummy leads and human leads at the top level?

That partner will not find the right continuation every time is true for active leads as for passive leads. No convincing arguments have been brought forward why passive leads are inherently worse in that respect.

Of course if someone will always assume the lead has to be from longest and strongest against all indications, this somebody will not defend well. Give him this book to read. Maybe it is time to make less simplistic rules past the beginner stage what to lead against 3NT.

If I lead an ace I will not find the right continuation every time.

True.

Do I find always the right continuation when I lead anything else the next time I have to take a decision? You must be joking and I am not known for suffering from an inferiority complex.

 

Rainer Herrmann

[whereagles]

well, I see there is no current trend towards switching gears w.r.t. opening leads default cards :D

 

(damn, that was convoluted)