Evaluation of hand play

[Wayne_LV]

What do you think?

[MrAce]

What do you think?

 

% 100 obviously http://www.bridgebase.com/forums/public/style_emoticons/default/cool.gif

[nige1]

IMO double-dummy analysis provides a good estimate of practical f2f results, provided human-players on both sides are of roughly equal standard. As declarer, I guess I might achieve 75%, on a good day.

 

100% is ridiculous, unless, like Mr Ace, you peek :)

[mikeh]

Well, if you're asking about how many tricks 'I' expect to make...it is at least 100% of the DD answer....often more because I use my advanced subliminal mind control techniques not only to play double dummy but also to induce horrific defensive errors.

 

However, my partners never seem to have the same occult abilities, so for them it is about 40%....which means on average, counting my overtricks, we get about 75%.

[Mbodell]

Are you counting on defense too? Including the opening lead?

 

I expect that I'll take the about GIB says or more about 75-80% of the time as declarer. I expect as defender I'll take the amount GIB says or more about 60-65% after the opening lead, but maybe as low as 35% if before the opening lead.

[Trinidad]

Is this a lesson in math?

 

GIB is a model predicting the number of tricks any player will take in f2f bridge. Sometimes GIB will favor the defense, sometimes it will favor declarer. Sometimes it will favor us, sometimes the opponents.

 

So, if you ask the "average" bridge player this question, on average GIB will be exactly correct. Part of the time you will get more, and an equal part of the time your opponents will get less. Since your opponents belong to the same population, the fraction of boards scoring more than the DD analysis is exactly the same as the fraction scoring less. This transforms your question to: What percentage of the hands does double dummy analysis predict the actual result exactly correct? The fraction of "tricks more or equal to double dummy" (let's call it ME) will then be ME=E+(1-E)/2= 0.5+E/2. (Where E is the fraction of the deals where the f2f result is exactly equal to the double dummy result.)

 

So, let's look at the fraction E. What would it be? I obviously don't know, but if I look at a typical traveller, I will see that when the board is played 20 times there will be 10 different results. Part of these are due to different contracts, but often there are differences in the number of tricks taken in the same denomination. I would estimate that on about 50% of the boards the actual number of tricks is exactly the same as the number of double dummy tricks, i.e. E=50%. Therefore the answer would be ME = 75%.

 

Obviously, there is a major flaw in this analysis. You didn't ask the average bridge player this question. You asked the BBO forum population this question. We all know that we are muuuccchhhhhh better bridge players than the rest. We will score more tricks than those other bozo's, so 95% (allowing for partner's stupidities) should be closer to the truth.

 

Rik

[rhm]

Research shows double dummy is a good predictor of the average actual number of ticks taken if either a tricky deal is played often in a specific denomination or over many deals.

However I do not know what the variance is.

Taking more tricks means you are better than the field and vice versa.

 

Also the research indicates, that you should take slightly more tricks in notrump partials and 3NT and you tend to take fewer tricks in slam contracts than GIB.

This is all explicable: In low level notrump contracts but including 3NT, the opening lead is often a decisive factor and nobody leads double dummy and in high level contracts, the defense has often far fewer options to go wrong than declarer.

 

So taking as many tricks as GIB would take does not necessarily mean you are good.

 

Rainer Herrmann

[helene_t]

I ususally compare my own results against three gibs to DD. I don't keep records but trying to eliminate positive recall bias I probably make too many tricks about 15-20% of the time, too few about 30-35%, so I end up with an overall accuracy of 50% as Rik said. I suppose this may be as much as 75% in the Bermuda Bowl and as little as 35% in my local club, maybe even less in the BBO main room.

 

It is well known that SD results tend to be biased in the favour of declarer, and that this bias gets weaker as the level of the contracts get up, with slam contracts showing the opposite bias: In a slam contract it is often so that declarer has to decide which finese to take and can get it wrong single dummy, while defenders have little influence on the result other than avoiding giving declarer a free finesse on the opening lead. But lower level contracts, especially when it goes 1NT-pass-pass-pass, are much more difficult to defend than to declare.

[gwnn]

58% of the statistics are made up on the spot, but estimates are even worse at 85%.

[helene_t]

58% of the statistics are made up on the spot, but estimates are even worse at 85%.

I just simulated 100000 bbf threads. It turned out that indeed 58% of the statistics quoted were made up, compared to 41% of bible quotes and 39% of bridge hands. So yes statistics has a pretty bad record.

[whereagles]

Research shows double dummy is a good predictor of the average actual number of ticks taken if either a tricky deal is played often in a specific denomination or over many deals.

 

Is there a scientific paper on this?

[rhm]

Is there a scientific paper on this?

To my knowledge scientific papers rarely deal with Bridge issues. I do not blame them.

The only one I know, which does, draws "scientific" conclusions about hand evaluation, I consider Bridge nonsense.

 

One old reference I have cited repeatedly is: http://crystalwebsite.tripod.com/double_dummy_accurate.htm

However, I would not claim that it is a "scientific paper".

 

Rainer Herrmann

[StevenG]

Has anybody researched how often the DD line equates to the best line? (I know everyone knows about dropping singleton kings and only taking finesses that work - but how often does this affect things?)

[hotShot]

Years ago I posted some data i produced comparing hand play with DD-Analysis.

My program calculated the DD-result after each played card. And counted how many tricks/deal where lost.

 

Beginner usually lose 1 trick/hand to DD. This would mean that if beginner play a deal declarer will usually make 1 trick more than the DD-result because he and both of his opponents will on average lose a trick.

The Beginner defense will usually lose 1 trick to the DD-Result.

 

Experts lose less than 1trick in 4 hands. Somebody with patience and knowledge in statistics can probably calculate a percentage out of this.

[helene_t]

Has anybody researched how often the DD line equates to the best line? (I know everyone knows about dropping singleton kings and only taking finesses that work - but how often does this affect things?)

Probably difficult to put a single number on this. What the "best line" is depends on how much information is available from the auction, opening lead etc. It is especially hard to establish a "best line" for the defense.

[whereagles]

hmmm... I see. Maybe it's high time we should get together and do some serious research for a journal.

[inquiry]

You can download results of players into dd-solver. Then click scorecard. From scorecard click "calculate par results." When that finishes save the file, then from the scorecard page, click the last menu item (DD bidding). A new window opens with a lot of statistics, including the % of time your play or defense was NOT double dummy, as well as the percent of time your opponents play and defense was not double dummy (that is par result independent, that is base upon the played contract).

 

Double dummy solver can be found at the bridge captain website.

 

 

[rhm]

You can download results of players into dd-solver. Then click scorecard. From scorecard click "calculate par results." When that finishes save the file, then from the scorecard page, click the last menu item (DD bidding). A new window opens with a lot of statistics, including the % of time , as well as the percent of time your opponents play and defense was not double dummy (that is par result independent, that is base upon the played contract).

 

Double dummy solver can be found at the bridge captain website.

Does it show "your play or defense was NOT double dummy" as you claim

 

or does it show

 

your result on the deal was not the same as double dummy?

 

And even if it does what you claim, what would that signify?

 

There are line of plays, which are simply inferior to the double dummy play and then there are plays, which are superior but which happen not to work on the actual layout.

 

Rainer Herrmann

[inquiry]

Does it show "your play or defense was NOT double dummy" as you claim

 

or does it show

 

your result on the deal was not the same as double dummy?

 

And even if it does what you claim, what would that signify?

 

There are line of plays, which are simply inferior to the double dummy play and then there are plays, which are superior but which happen not to work on the actual layout.

 

Rainer Herrmann

 

We all know that double dummy calculations will find the winning play on a given hand even if that play is WAY, WAY against the odds of being the "correct" way to play the hand. It is like the par result. If a slam requires four finesses and even suit breaks all around to make, if all that exist, the par result will be the slam that no one would ever be able to bid.

 

I only posted that the program did it in case anyone was curious as to how well their play match double dummy analysis. I should have added you can use DD-solver to filter your results by partnerships so you are only checking your results with a single partner, a small subset of partners, or all partners.

 

The results are presented like this....

http://3.bp.blogspot.com/-sn1195Bu4JE/U6rmwq3ONTI/AAAAAAAAAO8/nlUWDGIZ8fU/s1600/Capture555.JPG

 

 

As you can see, the top 2/3rds is related to the par contract, and gives things like if you miss game, miss slam, miss save, bid par contract, etc. The bottom 1/3 relates to the contracts actually played. So in this table the player involved defense was not double dummy 17.8% of the time (meaning it was double dummy or better about 82% of the time). The opponents double dummy defense was only 75% of the time. These were mostly expert players (JEC matches this past week).

 

Not sure what you can do with this, but there it is.

[barmar]

Does it show "your play or defense was NOT double dummy" as you claim

 

or does it show

 

your result on the deal was not the same as double dummy?

 

It's almost certainly the latter. On most hands, there are many different ways to achieve the same result. E.g. if you have Axx opposite KQx, you can take these 3 tricks in any order (assuming you don't need them as transportation to play in other suits at specific times, but even then there's no difference between playing the K and Q). How would you distinguish between "the same as double dummy" and "takes the same number of tricks as double dummy"?

 

But maybe what you're looking for is "declarer made a play that wasn't part of any double dummy best line, but then a defender misplayed as well so he got the trick back" (or vice versa).

[Zelandakh]

An interesting statistic that might be nice to add to that would be the percentage of individual plays that were worse than DD. So declarer has to choose 24 cards in a hand and analyse each of them to produce a percentage; the same for the 12 choices each defender makes. The advantage of this number would be that it is less affected by the level of the opposition. Not completely unaffected, since weaker opponents will make the best plays easier to find, but at least their direct mistakes do not get factored in.

[Wayne_LV]

Disappointing number of votes.

 

Time consuming as it was, I did a comparison of 500 IMP boards played this month by myself and 2 regular partners.

 

RESULTS:

 

WE DECLARED:

 

No. Boards............... 237

% Boards................ 47.4%

Average GiB DD..... 9.40 tricks

Average Actual......... 9.39 tricks

% = or Better......... 70.5%

IMP/Bd................... .58

 

 

WE DEFENDED:

 

No. Boards................ 263

% Boards................. 52.6%

Average GiB DD...... 3.71 tricks

Average Actual.......... 4.10 tricks

% = or Better........... 82.9%

IMP/Bd.................... .27

 

 

 

OVERALL:

 

No. Boards...............500

% Boards................100%

Average GiB DD..... 6.40 tricks

Average Actual......... 6.61 tricks

% = or Better.......... 77.0%

IMP/Bd................... .41

 

 

Is this good or bad? My feeling is there is room for improvement. I think a consistent average of 80-85% should be expected by Advanced pairs and real Experts should hit 90% +.

 

Note that the expected Average GiB DD is 6.50 or half the possible tricks.

[inquiry]

Wayne, since you are willing to publish your stats, I thought it would be useful to throw a larger sample up for you. This was not painful (actually easy) using the database method I described in the Excel trick threat and Double dummy solver.

 

I looked at 1538 hands you played on-line still available in myhands. I threw out 9 hands you played with random partners, leaving just your normal two partners for a total of 1529 hands. On these hands your side averaged +0.77 imps over 675 hands as declarer, but -0.001 imps over 689 hands as defenders (that I rounded up!). The average imps per hand over the total 0.38 imps over 1364 total imp hands.

 

The matchpoint results were similar. As declaring side, you averaged 0.52% over 82 hands, and as defender, 0.50 over 86 hands. The overall average was 0.51 over 168 hands.

 

But what you seem to be interested in is double dummy play. Here is the data from Double Dummy solver for all 1529 hands....

 

http://2.bp.blogspot.com/-5RRwJYYlB8k/U624qd80hlI/AAAAAAAAAPM/X2a__EcuOuE/s1600/Capture+wayne_lv.JPG

 

 

From this data, you see your partnership play was double dummy 74% of the time, and your double dummy defense was just slightly worse. Your opponents had very similar numbers. You seem to not be doubling often enough, and I would be concerned about how frequently you miss game (Since most hands are imps, you want to bid them aggressively) and this is supported by your very low "overbid" percentage. Your missed slam is also a bit high. Also your missed "sacrifices" seem a higher than normal (compare to your opponents). So I suggest a step up in your aggressiveness for little while and see if that improves your results (note I didn't look at any hands, so I may be full of crap).

 

All in all, your numbers here are not bad. I think between 75% and 80% for double dummy is about right. Your numbers for this and your opponents for that matter match the double dummy results for all the JEC opponents in the past month (see post 19) and most of them are at least high advanced or better (there are some exceptions). The worse your opponents, the higher your double dummy results will be because even if you screw up, they throw tricks back to you. The better your opponents, the worse your Double dummy results will be becase they are not giving you extra tricks as often to make up for your mistakes.