no trump hand evaluation methodology

[helene_t]

4-3-2-1 is actually surprisingly accurate when it comes to the decision to bid 3nt or not. I made a logistic regression analysis of the GIB DD database and found that the coefficients, when the outcome variable is "3nt makes DD", are very close to 4-3-2-1. Of course colaborating honours, tens and nines, and honours in long suits matter a bit. But devaluating aces (as in Banzai) or upvaluating aces (as in Viena points) will make it worse.

[msjennifer]

Playing a weak no trump,it is unwise to open a flat 4333 hand with 3Aces as 1NT.Same applies when two suits are wide open.Playing a standard system, and Red vs White ,one International opened 1 NT holding xxx,AKQx,AKJx,xx .All Pass.Going 3 down and minus 300 got the zero it deserved in the match point pairs.True, the opponents were making 3S. Now is it bad luck or unwise opening bid ?All others opened the hand 1 D and opponents played in spade partials..Using a rule of 20 or any other number does not help when opening a hand as 1 NT.A 1NT is passed out if partner holds a flat 6/7 HCP and goes down when a contract of two In a 4/4 suit is on ice if one had opened the hand 1C/D.

[Stephen Tu]

Playing a weak no trump,it is unwise to open a flat 4333 hand with 3Aces as 1NT.Same applies when two suits are wide open.Playing a standard system, and Red vs White ,one International opened 1 NT holding xxx,AKQx,AKJx,xx .All Pass.Going 3 down and minus 300 got the zero it deserved in the match point pairs.True, the opponents were making 3S. Now is it bad luck or unwise opening bid ?All others opened the hand 1 D and opponents played in spade partials..

Well the question that would come to my mind is how many of the others were strong NT pairs. Next time the auction might go 1d-p-1s-p, then what? 2H? 2nt? To me any choices can work out badly on a single hand, to really prove that it's unwise to open 1nt with 2 suits wide open you'd really have to do it statistically over a whole bunch of different hands.

[jogs]

Playing a weak no trump,it is unwise to open a flat 4333 hand with 3Aces as 1NT.

♠ A432 ♥ A32 ♦ A32 ♣ 432

Do you plan to pass? Or open 1♠ planning to pass 1NT, raise 2♥ or 2♦, and rebid 2♦ after 2♣.

[rhm]

4-3-2-1 is actually surprisingly accurate when it comes to the decision to bid 3nt or not. I made a logistic regression analysis of the GIB DD database and found that the coefficients, when the outcome variable is "3nt makes DD", are very close to 4-3-2-1. Of course colaborating honours, tens and nines, and honours in long suits matter a bit. But devaluating aces (as in Banzai) or upvaluating aces (as in Viena points) will make it worse.

I see some contradiction to Thomas Andrews research.

He claimed that the best coefficients for notrump were from Ace to Ten:

 

A=115

K=74

Q=43

J=23

T=10

 

He attaches also a double dummy value to the 9 (=4) and 8 (=2).

 

So according to Andrews Double Dummy analysis

 

An ace is equivalent to

 

Ace = 1.5 Kings = 2.7 queens = 5 jacks

 

Compare this to standard point count

 

Ace = 1.33 kings = 2 queens = 4 jacks

 

Of course the above coefficients are not practical, but they show that the ace is undervalued even at notrump and at least my double dummy studies tend to confirm this.

 

The fifth evaluator A=4 K=2.8 Q=1.8 J=1 T=0.4 is somewhere in between, more practical and considered more accurate than standard HCP. It also values the ace indirectly higher by reducing the value of king and queen and making room for the Ten within a total of 40 HCP per deal.

 

Rainer Herrmann

[Zelandakh]

A=115

K=74

Q=43

J=23

T=10

Let's normalise this by multiplying through by 3/74. Then:-

A = 4.66

K = 3

Q = 1.74

J = 0.93

T = 0.4

 

That gives a good basis for the idea of upgrading aces by (at least) half a point. What is missing from the analysis of the lower honours is whether they are with higher honours or alone in a suit. My opinion is that there is a big difference between the ten in AJTxx and Txxxx. Similarly, as Helene mentioned, being in a short suit devalues them further. So my method is to count the quacks as full value with a higher honour but 0.5 points less if alone; the Bergen evaluation that Ed gave, effectively discounting quacks by 0.3, is also fully justified. For a ten, counting it as 0.5 unless alone looks like it would work well (based on these numbers).

 

To the nearest half a point, the coefficients given are 4.5 - 3 - 1.5 - 1 - 0.5, which look rather familiar! Of course that 1.74 for a queen is annoying as it falls between 2 data points. We could double the values to give:-

A = 9.32

K = 6

Q = 3.49

J = 1.86

T = 0.81

 

To the nearest half integer, 9.5 - 6 - 3.5 - 2 - 1. Hmmm, too complicated? How about normalising to a K = 4 points:-

A = 6.22

K = 4

Q = 2.32

J = 1.24

T = 0.54

 

That is, A = 6, K = 4, Q = 2.5, J = 1, T = 0.5. That looks fairly good - perhaps in the future this is a method that could take off, even more so if we used Q = 2 and add half a point for an accompanied queen. What is very clear from these numbers is that the theoretical basis of Banzai points, which devalue aces relative to kings and quacks, is highly questionable. That is even more the case when one considers that a trump contract is still possible, as mentioned earlier by Rainer. I think the only reason for mentioning Banzai points in N/B is to warn readers not to use them!

 

The best way of evaluating though is more of an open question. I suspect that many different methodologies are used at expert level but almost all of them end up roughly in the same place. I suspect it would advance bridge theory somewhat for someone to do the work to implement a knowledge-based system, in the same way that doing this for positional analysis in chess created a revolution in computers and eventually in master play itself.

 

What I disagree with is jogs' assertion that we should be calculating directly in tricks. What should be obvious from the numbers in this post is that using points allows for far more flexibility in design while arriving at the same result. I am confident that this basis will remain for regular evaluation. Feeling constrained to use only tricks is useful only for creating something simple (and usually poor) for low level players or for those that do not understand the maths. It is almost always the case that a trick-based system can be improved by converting to points while retaining enough simplicity to be useful.

[helene_t]

As far as I can see on Thomas Andrew's site he gets the same results as I when he adresses the question of bidding 50% games. When he requires 40% he values the aces more aggresively, close to the numbers Rainer quoted.

 

There could be small discrepancies due to differences in subset selection ( I excluded hands with which we were obviously not aiming at a notrump contract, i.e. 9-card major suit fits and such but I don't remember the details). I also assumed that declarership was randomized, I think he does the same but obviously one could argue that a strong balanced hand should be assumed to be declarer most of the times as standard bidding systems tend to achieve that.

 

It is intriguing that aces should carry more weight when aiming at a 40% game than a 50% game. Intuitively I would think that the coefficients would be similar just the threshold different.

 

BTW the queens are valued less than two jacks. I suspect this is due to queens being slightly less valueable DD than SD because it doesn't matter that you are missing the queen as long as you have a 2-way finese since you always get the two-way finesse right. If my hypothesis is correct, bridgebrowser should value queens more favorably than GIB does. I have some bridgebrowser data, I will have a go with that next week.

[Zelandakh]

It is intriguing that aces should carry more weight when aiming at a 40% game than a 50% game. Intuitively I would think that the coefficients would be similar just the threshold different.

I suspect this is because very thin games are sometimes makeable only by holding up and hoping for a certain break whereas 50% games often have a power line or avoidance play to reduce the premium of this aspect of an ace. It would take a proper analysis of the data though to say for sure what is going on there.

[rhm]

Let's normalise this by multiplying through by 3/74. Then:-

A = 4.66

K = 3

Q = 1.74

J = 0.93

T = 0.4

Normalizing can be done in different ways, but I think the right way to normalize for comparison with standard HCP is to normalize so that the sum of the top 5 honors are equal to ten, so that the whole deal remains at 40 points.

 

This gives you the following results:

 

A=4.3

K=2.8

Q=1.6

J=0.9

T=0.4

 

Again you see aces are worth more and and quacks are worth less.

In suit contracts the spread is higher.

 

I do not know whether somebody has created a point count on this but it could be done fairly easily.

 

A=11

K=7

Q=4

J=2

T=1

 

This gives you 100 HCP per deck. A king plus a queen equals an ace, and a queen plus a jack and a ten equals a King.

Ballpark figures for 3NT would be 62+ HCP combined and 6NT would require 83+ HCP and 7NT 94+ HCP

A 15 to 17 notrump would be 37-42 HCP.

A 12-14 notrump would be 30-35 HCP

 

I am sure this is an improvement over current point count, but I am not sure by how much.

 

Rainer Herrmann

[Zelandakh]

A=11

K=7

Q=4

J=2

T=1

Exactly, another good option that illustrates the greater flexibility of working off of a point count rather than trying to force things into a basis of 13 (tricks).

[mycroft]

Liversidge, if you don't want to go down after 1NT AP, play a strong NT. Part of the benefit of 12-14 is the preemptive value of the bid, and that means you are going down sometimes (you're going down more than rarely with 15-17, too, by the way). I would want to see the scores and see why your down was a bad score - it is likely that partner had enough that if they come in, it's going down too, or you went -2 vul for 200. If the latter, oh well, it happens. Why didn't everyone else do it (as I know you're in Acol-land, so you should have a lot of company. Maybe everyone in your club plays variable?) If the former, then your opponents judged well or got lucky when they passed it out. They'll judge badly next time. The arguments are different in my world (where we're the only ones playing 12-14 in the entire room).

 

But the other reason you open 12-14 with three aces and spaces is that if you don't, you'll never catch up when partner opens. Occasionally (unless it's me, of course) your partner actually has something, and will never play you for 3 aces as a passed hand.

[Th0rss0n]

Going back to the point count, are not the K, Q, J & T all worth more when accompanied?

[dokoko]

There is another consideration, what will everybody else do ? Very few people will pass a flat 12 count these days, certainly not more, so you're going against the field if you do.

 

This sounds as if you shouldn't go against the field freely. But this is only true if

 

- the field's choice is superior or

- you have a definite edge in the play of the hand (this includes defensive play) or

- you prefer average minus to a 50-50 top-or-bottom shot.

 

Otherwise go against the field if it seems right.

[pescetom]

A=11

K=7

Q=4

J=2

T=1

That reflects the values cited here very precisely, errors of 0% 1% 3% 10% 4% respectively.

 

Compare to A=4 K=3 Q=2 J=1 T=0 which has errors of 0% 14% 25% 40% infinite% respectively.

 

But a good compromise might be to use A=3 K=2 Q=1 J=2/3 T=1/3 which has errors of 0% 3% 12% 10% 22% respectively.

That gives 7 points per suit which is easier to calculate mentally than 25 and still a lot more precise than 10. If counting the J and T in thirds is too much of a pain then probably someone can come up with a rule of thumb that assigns half-point range scores to various JT holdings.

[rmnka447]

Going back to the point count, are not the K, Q, J & T all worth more when accompanied?

 

Yes, they are, but how much more is a question for debate. Most of the really good players I know start from the basic point count and then use adjectives to characterize the hand -- "poor'. "bad", "nondescript', "decent", "good", "great", etc. They reflect the evaluation of the hand that goes past the point count and reflect the mental evaluation of plus and minus factors that affect hand value. These include positives such as honors working together, intermediates, intermediates working with honors, extra QTs, honors in long suits, or, negatives such as isolated honors (especially dangling Qs or Js), unguarded honors, weak long suits, lack of intermediates, intermediates not working with honors, lack of QTs.

 

Most often hands will have a combination of plus and minus factors that offset and the rating will be toward the middle of the scale of descriptions. Those hands will be bid normally. But sometimes hands will have lots of positive factors or negative factors that will predominate and the rating will be toward the top or bottom end of the scale. When they are toward the negative and near the bottom point count for an initial bid, the hands will be bid more pessimistically -- don't accept invitations -- or, in the extreme, choose a weaker initial bid. OTOH, when positive and near the top point count for an initial bid, the hands will be bid more aggressively -- invitations accepted, or, in the extreme, choose a stronger initial bid.

 

Maybe a few examples would help --

 

Consider ♠ AQx ♥ xx ♦ AQxx ♣ xxxx. You have 3 QT 12 HCP hand. The positives are the honors are working together and 3 QTs are more than usual for a 12 point hand. The negatives are no intermediates and one of the long suits is honorless. On balance, this is a good 12 point hand. If you're playing weak NTs (12-14), you ought to have little trouble opening 1 NT.

 

How about ♠ QJ ♥ AJx ♦ Qxxx ♣ Qxxx? This is 1 QT 12 point hand. But it has lots of negatives -- unguarded QJ, Q empty 4th in both long suits, no intermediates. This is a bad 12. If playing weak NTs (12-14), this is a hand you would likely pass rather than open 1 NT.

 

How about ♠ 1098 ♥ 109 ♦ AQ109 ♣ AQ109? This is again a 3 QT 12 point hand, but it has about as many positives you can get -- honors working in combination, lots of intermediates, and intermediates working with honors in the long suits. So it's a great 12.

 

Make it a little better ♠ K109 ♥ 109 ♦ AJ109 ♣ KQ109 -- 13 point 3 QT hand and you'd have little problem accepting any 2 NT invitation bid by partner.

 

Make it a little better ♠ A109 ♥ 109 ♦ AJ109 ♣ KQ109 -- 14 point 3 QT hand and you might even consider treating it like a 15-17 hand.

[bravejason]

Yes, they are, but how much more is a question for debate. Most of the really good players I know start from the basic point count and then use adjectives to characterize the hand -- "poor'. "bad", "nondescript', "decent", "good", "great", etc. They reflect the evaluation of the hand that goes past the point count and reflect the mental evaluation of plus and minus factors that affect hand value. These include positives such as honors working together, intermediates, intermediates working with honors, extra QTs, honors in long suits, or, negatives such as isolated honors (especially dangling Qs or Js), unguarded honors, weak long suits, lack of intermediates, intermediates not working with honors, lack of QTs.

 

Most often hands will have a combination of plus and minus factors that offset and the rating will be toward the middle of the scale of descriptions. Those hands will be bid normally. But sometimes hands will have lots of positive factors or negative factors that will predominate and the rating will be toward the top or bottom end of the scale. When they are toward the negative and near the bottom point count for an initial bid, the hands will be bid more pessimistically -- don't accept invitations -- or, in the extreme, choose a weaker initial bid. OTOH, when positive and near the top point count for an initial bid, the hands will be bid more aggressively -- invitations accepted, or, in the extreme, choose a stronger initial bid.

 

...

 

I agree with all of this.

 

Point count systems can be made very accurate, but then it becomes complicated and requires a lot of mental effort. I think is better to consider all the factors and then sum up the evaluation in a word or two. I like to use phrases like “minimal”, “invitational”, “full opening bid”, etc.

[0Filou]

I think that you will find the answers to most of the issues/questions raised here in the recently published book : Optimal Hand Evaluation (Master Point Press). In my opinion, It is the most comprehensive book ever published on hand evaluation and the most far-reaching and accurate.

[johnu]

I think that you will find the answers to most of the issues/questions raised here in the recently published book : Optimal Hand Evaluation (Master Point Press). In my opinion, It is the most comprehensive book ever published on hand evaluation and the most far-reaching and accurate.

Thank you for using this forum to advertise your book. You would do better if you wrote your own review on Amazon.com.

[nige1]

Normalizing can be done in different ways, but I think the right way to normalize for comparison with standard HCP is to normalize so that the sum of the top 5 honors are equal to ten, so that the whole deal remains at 40 points. This gives you the following results:

 A=4.3, K=2 .8, Q=1.6, J=0.9. T=0.4 

Again you see aces are worth more and and quacks are worth less. In suit contracts the spread is higher. I do not know whether somebody has created a point count on this but it could be done fairly easily.

 A=11, K=7. Q=4, J=2, T=1 

This gives you 100 HCP per deck. A king plus a queen equals an ace, and a queen plus a jack and a ten equals a King. Ballpark figures for 3NT would be 62+ HCP combined and 6NT would require 83+ HCP and 7NT 94+ HCP. A 15 to 17 notrump would be 37-42 HCP. A 12-14 notrump would be 30-35 HCP. I am sure this is an improvement over current point count, but I am not sure by how much.

That reflects the values cited here very precisely, errors of 0% 1% 3% 10% 4% respectively. Compare to

 A=4, K=3, Q=2, J=1, T=0 

which has errors of 0% 14% 25% 40% infinite% respectively. But a good compromise might be to use

 A=3, K=2, Q=1, J=2/3, T=1/3 

which has errors of 0% 3% 12% 10% 22% respectively. That gives 7 points per suit which is easier to calculate mentally than 25 and still a lot more precise than 10. If counting the J and T in thirds is too much of a pain then probably someone can come up with a rule of thumb that assigns half-point range scores to various JT holdings.

Brilliant work by Thomas Andrews, Helene_T, RHM, and Pescetom. My father's Winning Trick Count (WTC) evaluates

 A=1.5, K=1, Q=0.5 

with adjustments for knaves and tens. The numbers are trick estimates but, serendipitously, they agree with Pescetom's simple relative values :)

[MaxHayden]

Thank you for using this forum to advertise your book. You would do better if you wrote your own review on Amazon.com.

 

I will add to this further: I had been excited to write a both a forum review and an Amazon review of this book after I had time to test the results against the same type of datasets that were used in previous discussions about other point count systems. I am rapidly losing interest because of this behavior. The author is doing himself a huge disservice.

 

If he wants to promote the book, he should get involved in the community and actually provide people with helpful answers. Minimally he could link to the point-counting systems thread to help this person find more details. This generic spam is just going to result in no one taking him or the book seriously.