IMP scoring

[hotShot]

This is the score with crossIMPs

 

6 times 6♠ +1 1460 2.21 IMPs N/S

12 times 6♠= 1430 1.26 IMPs N/S

once 5♠+1 680 -11.63 IMPs N/S

once 7♠X-1 -200 -16.79 IMPs N/S

 

This is how it should be:

6 times 6♠ +1 1460 1 IMPs N/S

12 times 6♠= 1430 0 IMPs N/S

once 5♠+1 680 -13 IMPs N/S

once 7♠X-1 -200 -17 IMPs N/S

 

(As you can see Ben, the sum of the N/S scores is not 0 and i don't care.)

 

 

As we see the board is not very selective and the score does not split the field.

 

How can that be reached:

 

a) In a Butler way:

Eliminate 2 Scores at top and buttom, and calcucate the mean:

(12 * 1430 + 4*1460 ) /16 = 1438

score all pairs against this result.

 

b ) using the "median"

Meaning sorting the scores for NS by value and taking the one in the middle

(with an even number you middle the two closesed to that)

we have 20 scores here sorted scores 7-18 are 1430 so the median is 1430.

 

c) minimize the "absolute sum of IMPs"

At crossIMPs there were:

(6*2.21 + 12* 1.26 + 11.63 + 16.79) * 2 = 113,6

As it should be:

(6*1 + 12*0 + 13 + 17) * 2 = 72

 

Using crossIMPs in a tourney means that players making/defending the "par contract" will randomly get +/-IMP depending on which side the big scores/mistakes were made at another table.

Selection is done by seating not by bridge abilities.

 

The way i think it should be, seating is not (at least much less) selective.

 

This is my final post to this topic.

[inquiry]

This is my final post to this topic.

Well Good, Tim and I always like getting last word... :D

 

Ok, I will try one last time, for what it is worth. You first made a characterization above, looking at a tournment result where the net imps was minus SIX that this difference was because of some problem with cross imps. Even telling the world they could check any tournment and it would always be negative imps (as long as not survivor).

 

I pointed out why your example was flawed (average minuses for both NS pair added in, which accounted for the minus six imps).. Then I followed that up with two boards from a tournment you played in to illustrate the point that it really does add up to zero.

 

Even the TRAVELLER you present the NS scores do in fact sum up to ZERO despite your claim to otherwise (cut and past travelle rinto EXCEL and sum the column) (actualy it sums to minus 0.04 due to rounding errors). If you carried out the math to 9 decimal places, this becomes zero. It was this arguement you made, that the SCORE DOESN"T COME to ZERO that I was, and always have addressed.

 

Since you are going to stop posting, I will go ahead and address your other point. Which is on the hand given, the RESULT should be +1430 or maybe +1460... so haow come a +1430 is anything other than zero? Well this is a very hard standard to enforce. Lets say that 101 tables play a hand, and one pair bids 6S makes and all other bid 4S and make 6. What do you score in your example.. Should all the 4S bidders get minus 13 imps, since slam always make? I think with your example perhpas that is what should happen. But who judges 6sp is right.

 

The second arguement is cross imps cut the wrong way, so for the normal result you typically get a minus reesult. Well, that is out and out wrong. Let's take a simple example, you 6S example earlier.

 

6S is normal, so each NS pair here for the normal result got a small plus, but each EW on the same hand for the normal result got a small minus (after all, they did nothing wrong either). In fact, the normal result result in a net average (small plus for one side, small minus for the other). Sometimes you will bid a perfectly normal game, and make and get a small minus because a few agreesive people bid a horrible slam that should have no chance, but somehow made. Or you might get a small plus because some bid slam and went down and others couldn't find their way to game. I suggest you ignore these small differences, the extra work you propose and the extra calcuations hardly seem that would be in anyway better, and would take up valuable programmer time as well as server time. If you are this worried about the small differences, I recommend you play only matchpoints and team games... when I play "serious", I try to play teams becuase of issues that you are concerned about.

 

The views above are mine and mine alone as a user.. i have no idea how uday or fred feel about his issue.

 

Ben

[1eyedjack]

This is the score with crossIMPs

And next hand (all vul) the traveller shows:

 

19 tables 4S making exactly by N +620 to NS

1 table 5HX -2 by W for +500 to NS

 

Using your scoring the 19 tables score 0 Imps despite that at their tables the E/W pairs failed to find a profitable sacrifice.

 

Using cross imps the majority of the NS pairs get a small positive, as they deserve, having beaten par, and the EW get a corresponding negative, as they deserve.

 

You pays your money ...

 

I go with cross imps

[TimG]

This is the score with crossIMPs

 

6 times 6♠ +1 1460    2.21 IMPs N/S

12 times 6♠= 1430      1.26 IMPs N/S

once 5♠+1 680        -11.63 IMPs N/S

once 7♠X-1 -200      -16.79 IMPs N/S

 

This is how it should be:

6 times 6♠ +1 1460  1 IMPs N/S

12 times 6♠= 1430  0 IMPs N/S

once 5♠+1 680      -13 IMPs N/S

once 7♠X-1 -200    -17 IMPs N/S

 

(As you can see Ben, the sum of the N/S scores is not 0 and i don't care.)

 

Why should the EW pairs who defended 5♠ or 7♠ be rewarded? They did nothing but sit there and watch the opponents commit a blunder, yetthey get 13 or 17 IMPs.

 

a) In a Butler way:

    Eliminate 2 Scores at top and buttom, and calcucate the mean:

    (12 * 1430 + 4*1460 ) /16 = 1438

    score all pairs against this result.

 

IMPing against the mean is distorting due to the logarithmic nature of the IMP scale. Take a board where half the field scores +620 and the other half -100. All pairs will score +/- 6 IMPs = the average of the pushes and 720s (=12 IMPs). If you first take the mean = +260 and then IMP against that, all pairs will score +/- 8 IMPs (for their +/- 360 against the mean).

 

b ) using the "median"

    Meaning sorting the scores for NS by value and taking the one in the middle

    (with an even number you middle the two closesed to that)

    we have 20 scores here sorted scores 7-18 are 1430 so the median is 1430.

 

So, when there are 10 +620s and 9 -100s, everything is IMPed against +620 and when there are 9 +620s and 10 -100s, everything is IMPed against -100? I'd much rather average out this effect by IMPing against every pair and taking an average of that.

 

 

Using crossIMPs in a tourney means that players making/defending the "par contract" will randomly get +/-IMP depending on which side the big scores/mistakes were made at another table.

 

You can think of it as your expected score for that result given a large number of different teammates. Perhaps the reason there were more big scores/mistakes on one side of the score sheet is because the problems facing that pair were more difficult. As the number of replays increases, this ought to be closer and closer to the actuality.

 

 

This is my final post to this topic.

 

A sure sign that someone knows they have been beaten. :D

[hotShot]

Well i did not want to but...

 

I did not say that crossIMPs should be replaced.

 

I did say crossIMPs have a effect on the result, when NS and EW are mixed up in one result and extreme scores happen.

 

Matchpoints scoring has an effect on the result, as even a 10 point difference can give you a top.

 

In Totalpoint scoring there are effects an the result.

 

 

In a team match both tables either have the same result (0 IMP) or a different result leading to some IMPs.

So what i expect from an pair tourney with IMP scoring, is that a contract leads to 0 IMPs if played from most of the other players too.

At a team match the IMPs are not given compared to the best possible result, but to the (best) existing result.

 

I admit there are problems if you don't have a contract played by most of the others, as TimG points out.

 

a) In a Butler way:

    Eliminate 2 Scores at top and buttom, and calcucate the mean:

    (12 * 1430 + 4*1460 ) /16 = 1438

    score all pairs against this result.

 

 

IMPing against the mean is distorting due to the logarithmic nature of the IMP scale. Take a board where half the field scores +620 and the other half -100. All pairs will score +/- 6 IMPs = the average of the pushes and 720s (=12 IMPs). If you first take the mean = +260 and then IMP against that, all pairs will score +/- 8 IMPs (for their +/- 360 against the mean).

Well in a team match you would either have 12 or 0. I have no problem with 8 here, at least it is consitent.

 

 

b ) using the "median"

    Meaning sorting the scores for NS by value and taking the one in the middle

    (with an even number you middle the two closesed to that)

    we have 20 scores here sorted scores 7-18 are 1430 so the median is 1430.

 

 

So, when there are 10 +620s and 9 -100s, everything is IMPed against +620 and when there are 9 +620s and 10 -100s, everything is IMPed against -100? I'd much rather average out this effect by IMPing against every pair and taking an average of that.

Well i said the contract played by most of the others. I your example the majoraty is only a pair bigger than the minority.

 

Using crossIMPs at two boards, where all but one play a (small) slam, it will happen that my slam is only worth 1.2 IMPs (because someone did not bid it) while someone else's is worth 1.7 (because someone bid grandslam and fails with redouble). If most players make the slam, both should be avarded with 0 IMP.

 

 

I don't know how much CPU-Power your server has left, but Median and Butler scoring would require less than calculating crossIMPs. Sure the minimizing approach will cost (a little) more.

 

And if it would be implemented, if somebody wrote the perl script, i guess i can do it.

[1eyedjack]

a) In a Butler way:

    Eliminate 2 Scores at top and buttom, and calcucate the mean:

    (12 * 1430 + 4*1460 ) /16 = 1438

    score all pairs against this result.

 

b ) using the "median"

    Meaning sorting the scores for NS by value and taking the one in the middle

    (with an even number you middle the two closesed to that)

    we have 20 scores here sorted scores 7-18 are 1430 so the median is 1430.

 

c) minimize the "absolute sum of IMPs"

    At crossIMPs there were:

    (6*2.21 + 12* 1.26 +  11.63 + 16.79) * 2 = 113,6

    As it should be:

    (6*1 + 12*0 + 13 + 17) * 2 = 72

 

Using crossIMPs in a tourney means that players making/defending the "par contract" will randomly get +/-IMP depending on which side the big scores/mistakes were made at another table.

Selection is done by seating not by bridge abilities.

 

The way i think it should be, seating is not (at least much less) selective.

You appear to be suggesting three alternatives to cross imps, but despite that you have labelled them "a)", "b )", and "c)" it looks to me as though there are only two alternatives, with "c)" just being an extension of the consequences of "b )". Of the two alternatives (Butler and median) I am not sure which you are advocating as preferable, but I get the impression that it is the median method.

 

So, next hand you get:

 

10 tables in 4S making tick by N/S, +420

10 tables in 4S one off by N/S, -50.

 

Which would you treat as the median, against which to score all the results? +420 or -50? It makes a bit of a difference, I am sure you agree. I think from your post you are suggesting IMPing it against +235, but this seems just as random.

 

What if there were only 19 tables in total. You are the last table to finish and so far there are 9 tables making and 9 tables one off. You are a defender required to make a critical decision at trick 8. There are two possible layouts that might beat 4S, and you have nothing to go on. Which route you choose makes a massive difference to the scores at all 19 other tables as well as your own. Can this be an improvement on cross imps?

 

The flaws in Butler scoring have already been pointed out in

 

http://bridgebase.lunarpages.com/~bridge2/...indpost&p=20893

[EricK]

I suspect there is no perfect scoring system. But there may be one which is a bit better than the others.

 

On the other hand, I have never played any reasonably long session of bridge where I could honestly say that I lost or did badly because of the scoring. There are always plenty of hands where I could have played a bit better, and a few where I could have played much better. Once I have eliminated all those mistakes from my game, I shall return to the problem of determining the best scoring system.

 

Eric

[JRG]

Why should the EW pairs who defended 5♠ or 7♠ be rewarded? They did nothing but sit there and watch the opponents commit a blunder, yetthey get 13 or 17 IMPs.

Uh, I disagree.

 

They may have been very active in the auction. They may have pushed the opponents around; they may have accelerated the bidding and given the opponents a chance to go wrong. It is quite possible they were far from passive bystanders. They may have "earned" their good result.