Top level play Pavlicek data

[benlessard]

Its a follow up of

 

http://www.bridgebase.com/forums/topic/69070-interesting-stats-about-being-dealer/page__p__824684__fromsearch__1#entry824684

 

resume..

on 77 000 deals at high level play the raw scores of 1st and 3rd are not significantly higher than those of 2nd and 4th seat.

The score vs the par impsed give basically the same results.

 

 

By advantage I mostly mean do you expect to have a better score with the same hand if you are dealer vs the same hand but in 2nd seat. Assume you do all the 4 vul.

 

Basically if there is an advantage in all those case why are the overall results mostly even ? Are the light openings a costly proposition or its the weak nt hands that are costly ? What database query should we ask to explain the data results ?

[NickRW]

Advantage compared to what? Surely you're not asking is it an advantage to open 14-17 bal with 1NT compared to pass - or are you?

[billw55]

I am always happy when I get to bid first, so I put small advantage for all three. Bidder's game.

 

Of course, at teams the opponents will have the same situation so it should just be a wash. Not sure what to think overall.

[Zelandakh]

The score vs the par impsed give basically the same results.

Advantage compared to what?

Compared to par (see above).

 

I am always happy when I get to bid first, so I put small advantage for all three. Bidder's game.

 

Of course, at teams the opponents will have the same situation so it should just be a wash. Not sure what to think overall.

The OP is not talking about teams but rather in comparison with the par result. I am also surprised at this result and would expect 1st/3rd to have a (very) small advantage. How big is the sample size and margin of error? Maybe any advantage there is is being drowned out by noise from boards where there is no difference. Perhaps a different methodology should be used. Take all hands with a specific point count total from 1st seat and compare with hands from 2nd seat with the same total where Dealer opened - then check against par. Do the same for the other seats. That might help to spot a subtle trend in the data. Helene could probably come up with a better approach that could target in on potential discrepancies.

[helene_t]

I think Pavlicek made the best of the available data except that he failed to report confidence intervals. We had a discussion abou this some time ago and I made a back-of-the-envelop CI estimate that strongly suggested that the differences are not significant but it would be nice to have the raw data.

 

Ideally an experiment should be made with the same board being played in two team matches, rotated 90 degress relative to each other but such an experiment is probably not feasible. Maybe a group of club TDs could conspire for a few months to feed the same boards (except for rotation) to the dealer machines at their respective clubs, without telling the players that they are part of an experiment?

 

But of course it is more interesting to see how it works in high level competition. So I suppose we are stuck with Pavlicek's method. We might at some point get enough data in order to produce some low p-value but in any case we probably already know the conclusion: whichever direction the advantage goes to, the effect is almost certainly trivial.

[NickRW]

Compared to par (see above).

 

Sorry, but it isn't as simple as that. Ben says:

 

1) "The score vs the par impsed give basically the same results" (why bother asking then???)

 

2) "By advantage I mostly mean do you expect to have a better score with the same hand if you are dealer vs the same hand but in 2nd seat." (Well the same hand can't have been in both 1st and 2nd seat, so there is no comparison) Given that 1st hand will always pass, then, in terms of +/- ev, I'd rather have these hands in 2nd seat obviously, but I don't think that is what he is asking.

 

3) "Basically if there is an advantage in all those case why...". (Can't answer why because I still don't understand the comparison).

[Zelandakh]

Nick, try this: Given that you hold a specific hand, H, is your expected score the same as Dealer to sitting in second seat?

 

Going further, is E(H) different for different point ranges and is it affected by whether Dealer opens in front of you? This is a subject Ben has asked about before so I suspect his main interest is still in evaluating the practical effectiveness of opening light. That alone means that the question is not one of mere academic interest, esepcially for those of us that advocate light opening systems.

 

I would imagine that most of us would intuitively feel that for good hands (say, H > 12) there is an advantage to sitting in first seat or second after a pass as opposed to dealing with an opening bid in front of you. But Ben is saying that the statistics do not back that up. If true I still find that a surprising result. Is there anyone here that would have expected this? Given that surprise, I am still wondering if there is not something in the data that is hidden. That is why I wonder about finding a method that would fish for such a difference amidst all the noise. If we could come up with such a method, we might be able to, for example, use PK's database to obtain some meaningful results on the subject even without access to RP's data.

[awm]

It depends on what happens in 1st chair...

 

I'd expect that:

 

1. 5H and 12-14 probably a small advantage to be in 1st because opponents might bid 2S or something (giving me a problem) and this is probably more than I gain from times they help me in the play by locating points. 5S and 12-14 I'd expect to swing slightly the other way though, since preempting spades is harder.

 

2. Balanced strong, if opponents open at the two level it creates a problem, and a 1x opening may help them on lead. Probably an advantage to be in 1st.

 

3. Preempt, definite advantage in 1st.

 

I'd expect that when I have a normal PASS, its a real disadvantage to be in 1st chair, because my pass then gives opponents inferences they would not have from a 2nd chair pass over an opening.

[NickRW]

Nick, try this:...

 

Well, OK then. The pre-empt I want to hold in 1st seat quite strongly. The other two sound openings are less of an issue. Since 1st seat will often pass, we're left in a better position in 2nd seat, but set against that is that 1st may well open in which case the conventional view is that now we're behind.

 

As to Pavlicek's conclusions, if I were so inclined, I'd want to crunch the numbers myself, as like everybody else (and indeed RP himself if his comments are anything to go by), I am a little surprised.

 

Re light openings, I much prefer to have a system that permits unbalanced hands to be opened light. But balanced hands I can't see much (if any) advantage to sticking one's neck on the block early with anything less than a decent 12 count.

 

Nick

[benlessard]

Zel explanations are correct.

 

You have a close call between opening and not, there is a point on the curve where opening or not doesnt matter, both call will break even in the long run. This point depend on the overall system so it can be slightly different from one pair to another. When you open with a hand under this point you are leaking pts, its not because you dont get doubled that your opening costed you nothing...

 

from the opps perspective

 

1- sometimes its easier/better to overcall, make a Michael, takeout double or double and and bid than to open.

2- you play the hand with more informations

3- you can bid thin game knowing where the cards are going to be

4- When the light opener pass you have more informations about their pass since the pass is narrower.

 

I dont think you need duplicate comparaison for this, 77 000 from recent top level is a lot of results and should be a high enough sample.

 

Assume you play 100 deals of money bridge with duplicate scoring 2vs2, the data tell us that the advantage of being dealer is only +1.5503 per deal.

So the opps make you this offer; instead of dealer turning they are going to give you 77 pts to buy your 50 dealers seats. They will be dealer for 100 deal in a row but you start with +77. Im pretty sure you would laugh at the offer.

 

Im pretty sure the advantage should be higher than 1.55 pts per deal but why its not ?

 

About 1S or pass situation Pavlicek say this...

Opening light in spades, however, leaves little doubt as being a losing strategy. Why so? Opening light in any suit makes constructive bidding less accurate due to the wider range of opener’s hand. This is offset by the advantage in bidding first. When you hold the highest ranking suit, the advantage in bidding first is minimal (you can usually bid later) and the detriment to constructive bidding is maximal (fewer bids remain for exploration). At least that’s my take on it.

However if you at the number of deals its

1996-2012 Pass 258 692-665 = 50.99 104-97-57 = 51.36

So out of the 77 000 hands its only 27 IMPS in 258 deals so the too agressive 1S is doesnt explain why 1st and 3rd advantage in the data is smaller than expected.

 

But if we look at

 

Weak NT vs. Minor

 

Year............Winner....boards...IMP Percent..........WLT Percent

 

2009-2012.......Weak NT...444......825-801 = 50.74......165-173-106 = 49.10

2005-2012.......Weak NT...746......1377-1361 = 50.29....278-284-184 = 49.60

2001-2012.......Weak NT...975......1815-1692 = 51.75....368-357-250 = 50.56

1996-2012.......Weak NT...1162.....2217-2030 = 52.20....436-423-303 = 50.56

 

Strong NT vs. Minor

 

Year...........Winner....Boards....IMP Percent.........WLT Percent

2009-2012......Minor.....321.......636-519 = 55.06.....118-107-96 = 51.71

2005-2012......Strong NT.522.......929-921 = 50.22.....194-173-155 = 52.01

2001-2012......Minor.....663.......1234-1133 = 52.13...222-235-206 = 49.02

1996-2012......Minor.....793.......1491-1413 = 51.34...268-285-240 = 48.93

 

it tell me that 12-14 bal in 1m doesnt fare so well. Note that these are comparative result between 2 tables so its 2000x2 =4000 deals.

 

My gut feeling tell me its right to open light with a minor if this suggest a lead (I wanted to polled for this but there was a 3 polls limit). I don't think the top level field make too crazy preempts or that preempts are not effective. So I really think that is the 11-12 balanced hands opened in 1m that blew all the expected advantage of 1st and 3rd seat.

 

Also significant is that all vul its better to be 2nd and 4th than 1st and 3rd ! Its a bit like children football where the quarterback throw the ball so badly that you have more chances to score when your on defence than on offense.

[benlessard]

Also interesting is the

 

Open 1H vs. Open 4H

 

1996-2012.......Open 4H 32......122-60 = 67.03...17-9-6 = 62.50

 

Open 1S vs. Open 4S

 

1996-2012.......Open 4S 42......119-93 = 56.13...13-15-14 = 47.62

 

If one table opened 4M and the other 1M we can assume that the hand is somewhat close between 1M and 4M (its a strongish preempt or a flawed preempt like 7-4) normally bids/preempts that are right in the middle of their range will bring more benefits than bids/preempts that are on the borders of their range.

 

A clearcut 3S should bring better success on average than a 3S preempt that is close to 1S,2S or 4S. So a projection of this data imply that high level preempt are effective but of course the sample might be too small.

[NickRW]

...So I really think that is the 11-12 balanced hands opened in 1m that blew all the expected advantage of 1st and 3rd seat.

 

You make a powerful case for playing the weak NT and playing it soundly.

[awm]

These numbers don't mean what you appear to want them to mean.

 

First, the weak NT / strong NT numbers are ridiculously close. For example in 2005-2012, weak no-trump did better than 1m by 16 IMPs over 746 boards. That's noise, not a signal.

 

Second, these numbers are highly skewed by who the pairs are. Even in the top level events there are big discrepencies in skill level between the top teams and the weaker teams. For example, Fantoni and Nunes are a highly effective pair who open a weak no-trump on a wide range of shapes. They also go deep into a lot of elite events, so they are probably heavily represented in Pavlicek's data. Take them out of the stats and probably strong no-trump looks better (and likely by bigger margins than you have currently). Sure, it might be that some of Fantoni-Nunes success is because of their weak no-trump, but they are also considerably better declarers and defenders than even the top fields.

 

Third, I doubt there was really an effort to control for methods. How many of the 1m openings with a strong no-trump are 1♣ precision? How many with a weak no-trump are 1♣ Polish? What about pairs who use intermediate no-trump ranges like 14-16 and 13-15? What about upgrades and downgrades? These stats aren't really telling you anything useful about what no-trump range you should play.

[benlessard]

Just to make sure my claim is not about the best range of NT, my claim is that light balanced hand (11 NV or 12 vul)opened in 1m have a lower EV compared to the same hand in a later seat, this negate the dealer advantage that we would expect. I also think with less certitude that opening 11 or some 12 bal by an artificial 1m opening (or a non-lead suggestive 1m) is a losing proposition.

[awm]

Just to make sure my claim is not about the best range of NT, my claim is that light balanced hand (11 NV or 12 vul)opened in 1m have a lower EV compared to the same hand in a later seat, this negate the dealer advantage that we would expect. I also think with less certitude that opening 11 or some 12 bal by an artificial 1m opening (or a non-lead suggestive 1m) is a losing proposition.

 

You might be right, but I don't think you've given any evidence to support this.

 

It could easily be the "obvious passes" bringing down expected total points. Or it could be that some other opening is not as good as we think.

[benlessard]

It could easily be the "obvious passes" bringing down expected total points.

Possible but its seem unlikely to me. Its tough to know for sure.

 

Should this factor change how light we should open ? I dont think so.

 

The farther the pts we have are from our opening threshold the less costly it is.

 

If I usually open with 12 and have 7 pts once my 7 pts are showned I may still have an ace left so basically declarer got little information. With 7 pts but usually open 10 count and showned my 7 pts declarer know I dont have an ace or a king left. So basically the costly range is about 3 under the opening.

 

The difference in frequency between 7-9 pts and a 9-11 is significant however during the play its more likely to expose 7 out of 7 pts than 9 out of 9 pts. However if ive got 9 pts we are more likely to declare than if ive got 7 pts.

 

So i dont totally disagree that these pass may explain the lack of dealer advantage in the data but I dont think changing the opening threshold will significantly alter the CostxFrequency of these pass.

 

Note that its obvious passes in first seat compared to the same hand in other seat. So the same hand could fail to overcall and sometimes a failure to overcall give more info than a 1st seat pass.

[rhm]

So I really think that is the 11-12 balanced hands opened in 1m that blew all the expected advantage of 1st and 3rd seat.

How does this reconcile with Pavliceks other results at http://www.rpbridge.net/9x41.htm that opening one of a minor was a winner when the other room passed?

Many of these borderline hands were 11-12 balanced hands.

 

About 1S or pass situation Pavlicek say this...

 

Opening light in spades, however, leaves little doubt as being a losing strategy. Why so? Opening light in any suit makes constructive bidding less accurate due to the wider range of opener’s hand. This is offset by the advantage in bidding first. When you hold the highest ranking suit, the advantage in bidding first is minimal (you can usually bid later) and the detriment to constructive bidding is maximal (fewer bids remain for exploration). At least that’s my take on it.

Pure speculation with no real evidence. I have been brought up with the argument bidding 1♠ is attractive since the highest ranking suit has a preemptive effect on opponents.

Now we are told wait and see, there is no rush...

 

Rainer Herrmann

[NickRW]

Pure speculation with no real evidence. I have been brought up with the argument bidding 1♠ is attractive since the highest ranking suit has a preemptive effect on opponents.

Now we are told wait and see, there is no rush...

 

Well it is true that the reasons given are purely Pavlicek's speculation. However, the comparison between spades and the other suits remains quite striking and begs some sort of explanation even if RP didn't hit the nail on the head.

[benlessard]

How does this reconcile with Pavlicek other results at http://www.rpbridge.net/9x41.htm that opening one of a minor was a winner when the other room passed?

Many of these borderline hands were 11-12 balanced hands

 

Ive looked at the sample hands and most good result are from light 1m with suits and some serious misbids by the opposition over an art 1D. But its only 100 hands so i dont want to draw conclusions.

 

A good database query would be to look at just the 11 bal to see if they do better with 1m or pass. Even if it doesnt take into account the better minor VS art 1m it should be a good indicator.

[benlessard]

Pavlicek has responded to my query.

 

http://www.rpbridge.net/9x35.htm#5

 

Surprising stuff, its mostly all the balanced that get a weaker result in 1st seat not just the weakest ones.

raw scores

 

15-17 in 1st seat is worth 14pts less than in 2nd seat.

 

6card majors 1st seat is worth 28pts more than in 2nd seat.

 

Note these are not multi table comparaisons just the raw scores on many deals.

[PhilKing]

It's about what I thought.

 

Do I get a prize? ;)

[helene_t]

Any tactical implications of this? I know it's not a straightforward conclusion that we simply shouldn't open 12-17 balanced hands in first seat but .....

[NickRW]

Any tactical implications of this? I know it's not a straightforward conclusion that we simply shouldn't open 12-17 balanced hands in first seat but .....

 

I am not sure that it even begins to say that. Yes, it does suggest that balanced hands would be better opened if they were in 2nd seat compared to 1st - but so what - on any given deal with, say 12-14 bal in 1st seat, you have the hand you've got. The rules of the game preclude you suggesting to LHO that you swap seats! You primarily have to decide to open or not. This data (or at least the particular data to which Ben has recently referred), does not suggest that passing is better instead.

 

Or at least that is what I've understood.

 

Nick

[rhm]

Any tactical implications of this? I know it's not a straightforward conclusion that we simply shouldn't open 12-17 balanced hands in first seat but .....

That the modern trend to light openings is a winner for distributional hands, but dubious for balanced hands irrespective whether you play a strong club system or not?

Fantunes is a system, which adheres to this philosophy.

Statements like "we open all hands with 11 HCP" violate it.

Judgement still rules.

 

Rainer Herrmann

[cherdano]

Any tactical implications of this? I know it's not a straightforward conclusion that we simply shouldn't open 12-17 balanced hands in first seat but .....

My conclusion is to wonder about the standard deviation for the total score of a single bridge deal.

What did your back-of-the-envelope calculation lead to? Mine says the results for the various ranges of balanced hands don't even come close to being statistically significant. Of course, combined they might come closer, and if it's the result PhilKing expected, we should set a high Bayesian prior...