Interesting stats about being dealer

[benlessard]

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

 

Again very intersting stuff from Pavlicek website.

 

 

 

I think its safe to assume that preempting before opponents open is a pretty good advantage. I think its also safe to assume that hands with 5+M are better in 1st seat than after opponents actions.

 

If being first doesnt bring out a significant advantage over thousands of hands but some hand types have a clear advantage if they are dealer can we assume that some hand type that we open are at a clear disadvantage if they are in first seat ?? Is there a hole in this logic ?

[PhilKing]

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

 

Again very intersting stuff from Pavlicek website.

 

 

 

I think its safe to assume that preempting before opponents open is a pretty good advantage. I think its also safe to assume that hands with 5+M are better in 1st seat than after opponents actions.

 

If being first doesnt bring out a significant advantage over thousands of hands but some hand types have a clear advantage if they are dealer can we assume that some hand type that we open are at a clear disadvantage if they are in first seat ?? Is there a hole in this logic ?

 

In terms of calculating our results against par, one might assume there are types of hand that are a loser for opener whether we open them or not, and there ain't necessarily a great deal we can do about that. But there is a far more obvious way in which being first in hand can be a loser - for example, say we have a 2NT opener: hearing first hand open a nebulous club gives us far more information than if we were first to speak. Or perhaps we hold a weak 2 in hearts, but RHO oens 1♥, so we keep quiet and beat par ...

 

If we are the first person to have to divulge something about our hand, although that can steal space if it is their hand and helps us reach our best spot when it is ours, it is a double-edged sword. Think of it as being under-the-gun playing full-ring texas hold'-em.

 

I would not assume a closer analysis of the stats to lead to a Bocchiesque "big bang" in bidding theory, though doubtless a few fertile areas will emerge. Just don't expect that they will solve bridge.

 

 

 

 

 

 

Note for those who have not clicked the link - the Pavlicek stats show an overall loss against par for the side first to speak.

[rhm]

The statistics are remarkable.

It would be very interesting if we would have similar numbers for the time before 1996.

 

Over the last 50 years opening bids at high level tournaments have got about 2 HCP lighter.

My first Bridge book was "Bridge is an easy game" by Iain Macleod.

It explained the philosophy of ACOL with its "light" opening bids.

Recommendation in the book: Open all hands with 14 HCP.

 

One rationale for light opening bids (espoused at Bridge winners) for example by Kit Woolsey, who opens all 11 HCP hands:

 

"It is a fact that the opponents bid more accurately when they open the bidding than when you open the bidding. Consequently, it pays to open if the decision is close"

 

So how does this reconcile with these statistics?

 

Rainer Herrmann

[helene_t]

the Pavlicek stats show an overall loss against par for the side first to speak.

Huh? The first hands wins against par at imps but loses at bam.

 

I dislike his theoretical formula, though. First of all it is contraintutive if the weirdness depends on the vulnerability. So I think a parameterization with the vulnerability explicit would be more natural. Second, the imaginary number i factors out so it shouldn't be there. (Edit: my bad, i only factors out if s is an integer. Anyone who can estimate s from the data?)

 

Intuitively I would think that if the dealer doesn't have an advantage, our bidding systems suck.

 

But this is unfortunately just a statistical fluke. The standard error on the overall BAM win is sqrt(77411/4)/77411 = 0.0018 = 0.18% so the 95% CI is about 49.47% : 50.17%.

[PhilKing]

Huh? The first hands wins against par at imps but loses at bam.

 

I dislike his theoretical formula, though. First of all it is contraintutive if the weirdness depends on the vulnerability. So I think a parameterization with the vulnerability explicit would be more natural. Second, the imaginary number i factors out so it shouldn't be there. (Edit: my bad, i only factors out if s is an integer. Anyone who can estimate s from the data?)

 

Intuitively I would think that if the dealer doesn't have an advantage, our bidding systems suck.

 

But this is unfortunately just a statistical fluke. The standard error on the overall BAM win is sqrt(77411/4)/77411 = 0.0018 = 0.18% so the 95% CI is about 49.47% : 50.17%.

 

Ah yes. I was looking at the last line in the table, which is of no relevance whatsoever. Line 5 is the relevant one.

[NickRW]

Double post

[NickRW]

Intuitively I would think that if the dealer doesn't have an advantage, our bidding systems suck.

 

Could easily be the case. :) I can think of "assumed logic" underlying a lot of peoples' attempts at creating a suite of opening bids, but not a whole bunch of anything that amounts to real science.

[helene_t]

I was interested in what this freakness parameter s so I evaluated it numerically (sorry, just using silly trapez integrals. Csaba can do this better):

 

 

Now I don't know whether the DA in the formular is MP, IMP or total points but if we take it that it is total points, then the overall freakness is about -1.3. For none vul and dealer vul it will be about -1.7 and -2.2, repectively, but the negative DA values are not consistent with the model.

 

What is the interpretation of the freakness? Presumably, in clubs where people often open out of turn, the DA will be small corresponding to a freakness around 0.6, but what other things can factor into it?

[jogs]

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

 

Again very intersting stuff from Pavlicek website.

 

 

 

I think its safe to assume that preempting before opponents open is a pretty good advantage. I think its also safe to assume that hands with 5+M are better in 1st seat than after opponents actions.

 

If being first doesnt bring out a significant advantage over thousands of hands but some hand types have a clear advantage if they are dealer can we assume that some hand type that we open are at a clear disadvantage if they are in first seat ?? Is there a hole in this logic ?

 

Pavlicek doesn't give the variance or standard deviation of his results. Doesn't give us access to the raw data. Dealers vulnerable are either opening too often or just opening the wrong hands. It would be shocking if with both vul, dealer were actually at a disadvantage.

[PhilKing]

People are drawing an erroneous conclusion from the data.

 

The issue for a player first in hand is not whether he can get above par, it is whether an action will get closer to that goal than an alternative. You simply cannot conclude that first in hand opening bid theory is wrong.

 

A simple example: we open 1NT in first seat at red with a balance 15 count, get doubled and go for 1100. Does that mean we should change the system? Of course not. I'm not claiming the data is going to be full of swings like that, but there are many more mundane situations where simply opening the bidding is going to hurt is when we end up defending. It's just a question of information leakage. And that's not just when we open, its when we pass. Think of all the times you play guess a queen right just because you know someone has not opened.

 

Now if we broke the stats down to just include first in hand preempts, I would be very surprised if the opponents beat par.

[jogs]

 

 

A simple example: we open 1NT in first seat at red with a balance 15 count, get doubled and go for 1100. Does that mean we should change the system? Of course not. I'm not claiming the data is going to be full of swings like that, but there are many more mundane situations where simply opening the bidding is going to hurt is when we end up defending. It's just a question of information leakage. And that's not just when we open, its when we pass. Think of all the times you play guess a queen right just because you know someone has not opened.

 

 

It's all about frequency. If 1NTX goes -500 against air too often, you should consider not opening 1NT with 15 flat.

 

We can't draw firm conclusions from Pavlicek's stats. Players may be pre-empting too often vul. It would be nice to have access to the raw data. Design our own test.

[nige1]

http://www.rpbridge.net/g/da.gif

 

IMO Pavlicek's formula is intended as a joke :).

[jogs]

Pavlicek does have a valid point on boards played. It should be 16 board segments, not 15. After 64 boards each of the four possible vulnerabilities occurs 16 times.

[jogs]

Pavlicek needs a study of the advantages(or disadvantages) of opening. Study the boards where one dealer opened and the other dealer passed. Which group won imps? Those who opened or those who passed?

[benlessard]

http://www.rpbridge.net/9x41.htm

 

http://www.rpbridge.net/rpme.htm

 

Its the same 77,000 but it lead that opening 1m/1H instead of pass is +ev while in 1S/vs pass pass is the winner.

 

However i dont think the number of hands is high enough.

[jogs]

http://www.rpbridge.net/9x41.htm

 

http://www.rpbridge.net/rpme.htm

 

Its the same 77,000 but it lead that opening 1m/1H instead of pass is +ev while in 1S/vs pass pass is the winner.

 

However i dont think the number of hands is high enough.

 

There needs to be std dev shown. That's how one knows whether the sample size is sufficient.

Or at least one can determine the reliability of the conclusions.

[jogs]

Didn't like the way Pavlicek constructed his tables.

Reconstructed one table by 4 year segments.

Keep W(in) as when 1S won.

 

 

                            Open 1S vs. Open 1NT
Year		Winner		Boards	IMP Percent	WLT Percent
2009-2012	Open 1S		85	191-109 = 63.67	36-29-20 = 54.12
2005-2012	Open 1S		146	317-188 = 62.77	60-49-37 = 53.77
2001-2012	Open 1S		165	345-215 = 61.61	65-55-45 = 53.03
1996-2012	Open 1S		183	377-265 = 58.72	71-63-49 = 52.19

 

                          Open 1S vs. Open 1NT
Year		Winner		Boards	IMP Percent	WLT Percent
2009-2012	Open 1S		85	191-109 = 63.67	36-29-20 = 54.12
2005-2008	Open 1S		61	126-079 = 61.46	24-20-17 = 54.55
2001-2004	Open 1NT	19	026-027 = 49.06	05-06-08 = 45.45
1996-2000	Open 1NT	18	032-050 = 39.02	06-08-04 = 42.86

[helene_t]

The standard error on the overall BAM win is sqrt(77411/4)/77411 = 0.0018 = 0.18% so the 95% CI is about 49.47% : 50.17%.

Actually this is assuming that the BAM scores are all wins or loses but obviously there will be a significant amount of ties also. I can't calculate the exact CI without knowing how often that occurs. Anyway, it is probably safe to say that the reported findings are not statistically significant.

[jogs]

The standard error on the overall BAM win is sqrt(77411/4)/77411 = 0.0018 = 0.18% so the 95% CI is about 49.47% : 50.17%.

 

No, it is not that low. You are confusing the variance of the estimate of the mean with the population variance.

The population variance is unchanged by the size of the sample.

[cherdano]

No, it is not that low. You are confusing the variance of the estimate of the mean with the population variance.

The population variance is unchanged by the size of the sample.

LOL

[jogs]

http://www.macroption.com/population-sample-variance-standard-deviation/

 

Here's the link. Shows how to use sample variance to estimate population variance.

-----------------------

Take boards from BBO minis. Choose a matchpoint mini with over

50 tables. The boards are scored 0 to 100%. The std dev of a board

is approximately 28%. BAM is mps with a 1 top.

For BAM the std dev is approximately 0.28 when 50% of the boards

are pushes. The std dev is above 0.32 when about one-third of the

boards are pushes. For the std dev to drop under 0.25 there needs

to be over 60% pushes.

[cherdano]

jogs, in case you really don't get it: the LOL'ed was one of the most well-deserved in BBF history, because

- you explained something blatantly obvious, and

- you didn't realize something else that's blatantly obvious, namely that the variance for the estimate of the mean is what is relevant for Helene's point.

 

If you delete your posts (which I'd do in your position), I will do you a favour by deleting mine which quoted yours.

[benlessard]

Was That Sarcasm?

--no

Was That Sarcasm?

--yes

Was That Sarcasm?