Flat Hands

[Douglas]

I can remember seeing a statement like "hand-dealt deals have more flat hands than computer-dealt ones" in this group. I also remember asking "what is a flat hand?" in this same group, and receiving stony silence. Due to my recent discoveries related to bridge deal hand-types, I now notice I have a point of view about this general subject.

My limited experience (compared, I'm sure, to most of you) causes me to think of flat hands mostly when considering bidding an opening no trump. Personally, I limit myself to three hand types: 4333, 4432, and 5332. I understand some players are more daring with their distributional possibilities.

If I am to take a flat hand literally, 4333 is the flattest possible, and if I were conservative minded, I could limit myself to thinking of it as "the" only flat hand.

If so, I am sorry to tell you for both hand, and computer-dealt, deals, there is no particular excess quantities of 4333 hands observable over time. It tends to be very ordinary in its variation from its theoretical expected value.

4432 and 5332 are entirely different. In hand-dealt deals they are regularly over-abundant. That is the major hallmark that sets these two kinds of dealings apart statistically.

If you are one who includes 5422, or even 6322, into your no trump opening calculation, you have a broader definition of a "flat hand." These additional hand-types are ordinary in their random distribution also, and would makes all these hand-types together more ordinary as a grouping of "flat hands."

Douglas

[Barry Margolin]

Post by Douglas
I can remember seeing a statement like "hand-dealt deals have more flat hands
than computer-dealt ones" in this group. I also remember asking "what is a
flat hand?" in this same group, and receiving stony silence. Due to my recent
discoveries related to bridge deal hand-types, I now notice I have a point of
view about this general subject.

I think the claim is that human-dealt hands tend to be flatter. So you
get fewer singletons, voids, and 6+ card suits.

Post by Douglas
My limited experience (compared, I'm sure, to most of you) causes me to think
of flat hands mostly when considering bidding an opening no trump.
Personally, I limit myself to three hand types: 4333, 4432, and 5332. I
understand some players are more daring with their distributional
possibilities.
If I am to take a flat hand literally, 4333 is the flattest possible, and if
I were conservative minded, I could limit myself to thinking of it as "the"
only flat hand.
If so, I am sorry to tell you for both hand, and computer-dealt, deals, there
is no particular excess quantities of 4333 hands observable over time. It
tends to be very ordinary in its variation from its theoretical expected
value.

Where do you get your data?

Post by Douglas
4432 and 5332 are entirely different. In hand-dealt deals they are regularly
over-abundant. That is the major hallmark that sets these two kinds of
dealings apart statistically.
If you are one who includes 5422, or even 6322, into your no trump opening
calculation, you have a broader definition of a "flat hand." These additional
hand-types are ordinary in their random distribution also, and would makes
all these hand-types together more ordinary as a grouping of "flat hands."

Note that the a priori odds of 4432 distribution is more than twice that
of 4333. The raw percentages are that 4432 is 21.6%, 5332 is 15.5%, and
4333 is 10.5%.

So since 4333 is relatively infrequent to begin with, a difference due
to dealing method would be less noticeable.

[Douglas]

I think the claim is that human-dealt hands tend to be flatter. So you get fewer singletons, voids, and 6+ card suits.

How interesting you say that. I reported this precise result from a large number of hand-dealt deals some years ago in this group, and was told I did not understand anything about probability; something about my not reading the right books, or the right number of books, about the subject by someone named Phillips as I remember. As I also remember, my database of actual hand-dealt deals then was about 2,600.

Where do you get your data?

That same database eventually grew to more than 5,500 deals. I eventually tossed all but almost 500 deals I personally supervised being dealt, and recorded.

I never trusted my outside procured data very much, and I did not think I would ever find a practical way to large-scale analyze it. It was all in PDF format, and even now is not able to be programmed for statistical results. Everything has to be hand transcribed, which is tedious and error prone.

Note that the a priori odds of 4432 distribution is more than twice that of 4333. The raw percentages are that 4432 is 21.6%, 5332 is 15.5%, and 4333 is 10.5%.
So since 4333 is relatively infrequent to begin with, a difference due to dealing method would be less noticeable.

So, let's see if this factually true. My absolute minimum size analysis unit of interest is currently 200 deals, which means 800 trial hands. 10.5% of that is 84 as the expected mean for 4333 hands. That seems to me to be a more than adequate number of degrees of freedom to accommodate virtually any possible variate returned.

Since each particular variate returned can be turned into a p-value, and from there into a standard deviation from expected mean, any variate from any category, big or small, is directly comparable in generally accepted standard units. This may be the main strength of modern basic statistics thought and practice.

Douglas

[jogs]

Post by Douglas

I think the claim is that human-dealt hands tend to be flatter. So you get fewer singletons, voids, and 6+ card suits.

How interesting you say that. I reported this precise result from a large number of hand-dealt deals some years ago in this group, and was told I did not understand anything about probability; something about my not reading the right books, or the right number of books, about the subject by someone named Phillips as I remember. As I also remember, my database of actual hand-dealt deals then was about 2,600.

Where do you get your data?

That same database eventually grew to more than 5,500 deals. I eventually tossed all but almost 500 deals I personally supervised being dealt, and recorded.
I never trusted my outside procured data very much, and I did not think I would ever find a practical way to large-scale analyze it. It was all in PDF format, and even now is not able to be programmed for statistical results. Everything has to be hand transcribed, which is tedious and error prone.

Note that the a priori odds of 4432 distribution is more than twice that of 4333. The raw percentages are that 4432 is 21.6%, 5332 is 15.5%, and 4333 is 10.5%.
So since 4333 is relatively infrequent to begin with, a difference due to dealing method would be less noticeable.

So, let's see if this factually true. My absolute minimum size analysis unit of interest is currently 200 deals, which means 800 trial hands. 10.5% of that is 84 as the expected mean for 4333 hands. That seems to me to be a more than adequate number of degrees of freedom to accommodate virtually any possible variate returned.
Since each particular variate returned can be turned into a p-value, and from there into a standard deviation from expected mean, any variate from any category, big or small, is directly comparable in generally accepted standard units. This may be the main strength of modern basic statistics thought and practice.
Douglas

That's your background in statistics? In poker simulations there are 1 billions deals(takes under 10 seconds 10 years ago). Your sample size is extremely small.

[Douglas]

Post by jogs
Your sample size is extremely small.

You noticed!

But it is just as important to notice it is my absolute minimum. I arrived at it through extensive experience, and I remain somewhat amazed by how small it is. Let me explain a bit.

Previously I thought suit length was the basic unit in bridge dealing analysis. It seemed so logical to me. But relatively recently I "accidentally" did an analysis of reported hand-types, got unexpectedly good results, and finally noticed the factor I had not paid attention to previously. Somewhere I read statistics is about probability and distributions. I am more convinced about that now.

It is the shape of the bridge hand-type distribution which makes the measurement I use here so distinct at a mere 200 deals possible. When I used suit lengths as my analysis unit of interest, it took at least 600 deals to achieve the beginning of any obvious statistical distinction. As should be obvious to the interested person, each 4 suit lengths added together make up one hand-type. Who knew such a subtle distinguishment could make such an analytic difference.

Enough talking about it. Here are real numbers from an experiment I recommended to another person in this group.

These first 250 deals are from the Reno NABC this past Spring (Set NABC161-03-16-100 et al.):

48 0.382
105 0.738
155 0.698
205 0.674
262 0.848
308 0.730
356 0.661
399 0.486
456 0.643
499 0.487

My experience in postings to this group is tables do not work well. The first column is the accumulated totals from each 25 deals. The second column is their associated cumulative binomial p-values. If you are used to reading this kind of information, you should see the p-values "converging" toward 0.50, the expected mean. I would describe this particular convergence as fairly rapid. Which, in my experience, likely makes this a typical result from a usual congruential number generator.

This second result table is from my first 200 supervised actual mixed up and hand dealt bridge deals database. There are only 200 deals because by then the distinction is clearly different, and I have no need to continue.

58 0.933
106 0.782
154 0.657
210 0.829
268 0.941
329 0.990
378 0.981
433 0.989

This is classic "divergence" in action.

Douglas

[Steve Willner]

Post by Douglas
The first column is the accumulated totals from each 25 deals.

Totals of what?

Post by Douglas
This second result table is from my first 200 supervised actual mixed
up and hand dealt bridge deals database.

So these are hand-shuffled deals, for which you recorded "column 1?"

Post by Douglas
58 0.933
106 0.782
154 0.657
210 0.829
268 0.941
329 0.990
378 0.981
433 0.989

[p***@infi.net]

Post by Douglas
I can remember seeing a statement like "hand-dealt deals have more flat hands than computer-dealt ones" in this group. I also remember asking "what is a flat hand?" in this same group, and receiving stony silence. Due to my recent discoveries related to bridge deal hand-types, I now notice I have a point of view about this general subject.
My limited experience (compared, I'm sure, to most of you) causes me to think of flat hands mostly when considering bidding an opening no trump. Personally, I limit myself to three hand types: 4333, 4432, and 5332. I understand some players are more daring with their distributional possibilities.
If I am to take a flat hand literally, 4333 is the flattest possible, and if I were conservative minded, I could limit myself to thinking of it as "the" only flat hand.
If so, I am sorry to tell you for both hand, and computer-dealt, deals, there is no particular excess quantities of 4333 hands observable over time. It tends to be very ordinary in its variation from its theoretical expected value.
4432 and 5332 are entirely different. In hand-dealt deals they are regularly over-abundant. That is the major hallmark that sets these two kinds of dealings apart statistically.
If you are one who includes 5422, or even 6322, into your no trump opening calculation, you have a broader definition of a "flat hand." These additional hand-types are ordinary in their random distribution also, and would makes all these hand-types together more ordinary as a grouping of "flat hands."
Douglas

I originally learned that "flat" meant 4333, "balanced" meant no singleton or void, no more than one doubleton, and "semi-balanced" meant 5433 or 6322. Then I found people using "flat" to refer to 4432 and 5332 shapes.

[t***@att.net]

What is your second column? It's not clear to me what's there.

[peter cheung]

In theory there is no difference between hand deal or computer deal if they are all fair deals.
In practice some computer deals actually throw away some very uninteresting deals because that will make the game more interesting. deals that are all 4333 and 10 hcp deals are not used in actual games sometimes.

[Douglas Newlands]

Post by peter cheung
In theory there is no difference between hand deal or computer deal if they are all fair deals.
In practice some computer deals actually throw away some very uninteresting deals because that will make the game more interesting. deals that are all 4333 and 10 hcp deals are not used in actual games sometimes.

That's illegal.

doug

[Barry Margolin]

Post by Douglas Newlands

Post by peter cheung
In theory there is no difference between hand deal or computer deal if they
are all fair deals.
In practice some computer deals actually throw away some very uninteresting
deals because that will make the game more interesting. deals that are all
4333 and 10 hcp deals are not used in actual games sometimes.

That's illegal.
doug

It's sometimes done in special games, like "goulash" tournaments on BBO.
But it should never be done in regular games/tournaments, especially if
the players aren't forewarned about it.

[Player]

Forewarning has nothing to do with it. It is illegal. I would be mightily annoyed of this happened in a game in which I played. I read posters here passing hands which I would open and on which consequently get a huge advantage.

[Barry Margolin]

Post by Player
Forewarning has nothing to do with it. It is illegal. I would be mightily
annoyed of this happened in a game in which I played. I read posters here
passing hands which I would open and on which consequently get a huge
advantage.

I thought there was some provision in the Laws for special games, where
the sponsoring organization can use cooked deals, but I can't find it.
But many clubs have a tradition like "Holiday Goulash Party".

If you don't want to play in those games, don't go to the party. That's
the point of announcing it ahead of time.

[Lorne Anderson]

Post by Barry Margolin

Post by Player
Forewarning has nothing to do with it. It is illegal. I would be mightily
annoyed of this happened in a game in which I played. I read posters here
passing hands which I would open and on which consequently get a huge
advantage.

I thought there was some provision in the Laws for special games, where
the sponsoring organization can use cooked deals, but I can't find it.
But many clubs have a tradition like "Holiday Goulash Party".
If you don't want to play in those games, don't go to the party. That's
the point of announcing it ahead of time.

In the UK the rule is that you can play whatever you like but if it does
not conform to the rules then you can't issue master points so a holiday
goulash is fine as long as the results are not sent in for master point
alloction.

If you want some real fun at Xmas then try reverse bridge. Bidding
follows normal rules except you are contracting for the minimum tricks
you will lose. ie in a contract of 1D you make it if you win 6 tricks
or less. If you win 3 tricks it is equivlent to 1D+3 (ie 130 points).
If you bid 4H you must win 3 tricks or less with hearts as trumps to get
your game. A grand needs you to win no tricks so the oppo will try and
beat it by leading a 2 and hoping partner has the 3 or 4 so declarer has
to win.

[Barry Margolin]

Post by Lorne Anderson

Post by Barry Margolin

Post by Player
Forewarning has nothing to do with it. It is illegal. I would be mightily
annoyed of this happened in a game in which I played. I read posters here
passing hands which I would open and on which consequently get a huge
advantage.

I thought there was some provision in the Laws for special games, where
the sponsoring organization can use cooked deals, but I can't find it.
But many clubs have a tradition like "Holiday Goulash Party".
If you don't want to play in those games, don't go to the party. That's
the point of announcing it ahead of time.

In the UK the rule is that you can play whatever you like but if it does
not conform to the rules then you can't issue master points so a holiday
goulash is fine as long as the results are not sent in for master point
alloction.

I assume the same is true in ACBL. These types of games are just for
fun, not serious competitions.

[p***@infi.net]

Post by Douglas Newlands

Post by peter cheung
In theory there is no difference between hand deal or computer deal if they are all fair deals.
In practice some computer deals actually throw away some very uninteresting deals because that will make the game more interesting. deals that are all 4333 and 10 hcp deals are not used in actual games sometimes.

That's illegal.
doug

If the option to tweak "freakness" or whatever exists on the dealing software, I'm sure someone somewhere will do it, thinking "what's the harm." In fact, I've seen so many hands with two or three voids around the table lately I should perhaps ask whether this is being done at our club. I know a director stacked a hand once in the days before computer dealing; I told him that was completely unacceptable. And I suspect that certain players dealt dealt goulash hands back then as well.

[Robert Chance]

ISTR that one of the London rubber bridge clubs (possibly the Portland?) used to have an agreement that all undoubled 1-level contracts were automatically conceded without a card being played, and the cards would then be goulashed.

Added an extra level of unpredictability to the game... and also meant that if you were playing for a higher stake than you could really afford, you had to intervene over 1NT on hands that you normally wouldn't.

[Barry Margolin]

Post by Robert Chance
ISTR that one of the London rubber bridge clubs (possibly the Portland?) used
to have an agreement that all undoubled 1-level contracts were automatically
conceded without a card being played, and the cards would then be goulashed.

A common tradition in rubber bridge is to goulash hands that are passed
out.

Of course, none of this makes sense in duplicate bridge.

[p***@gmail.com]

Post by peter cheung
In theory there is no difference between hand deal or computer deal if they are all fair deals.

By definition.

But hand deals are never perfectly random. I think there need to be a minimum of three shuffles to get good results, and also think that most do not do that.

[jogs]

Post by peter cheung
In theory there is no difference between hand deal or computer deal if they are all fair deals.
In practice some computer deals actually throw away some very uninteresting deals because that will make the game more interesting. deals that are all 4333 and 10 hcp deals are not used in actual games sometimes.

Isn't that a less than one chance in one million? Maybe less than one chance in 10 million.

[Andrew B]

Post by jogs

Post by peter cheung
In theory there is no difference between hand deal or computer deal if they are all fair deals.
In practice some computer deals actually throw away some very uninteresting deals because that will make the game more interesting. deals that are all 4333 and 10 hcp deals are not used in actual games sometimes.

Isn't that a less than one chance in one million? Maybe less than one chance in 10 million.

FWIW, I checked this by dealing 10 million pseudo-random hands, and 10
of them were all 4333 and 10 HCP. (About 250 were all 4333 and < 12 HCP).