Tossing out Scores.

[matmat]

Currently all of the 16 times a board is played the scores are calculated into the average. Would it be possible to have it so that the two extremal scores are thrown out in the calculation (toward IMP or MP scores)?

[Gerardo]

Note: Talking for myself, not BBO.

 

No, it wouldn't.

 

In MP it clearly makes no sense, because how MPs get calculated (you compare against each other result, get a 1/0.5/0 if your score is better/equal/worse than the other, then (usually) your final result is converted to a percentage, so 10 MPs comparing against 15 other tables gets a 66.6%

 

Now, BBO uses CrossIMPs, which are calculated exactly the same way as MPs do, but you get the IMPed difference in each comparison instead of 1/0.5/0. Then (and NOT before), the sum of all your comparisons is (optionally*) divided by the number of comparisons.

 

IMO, this is the best method. I also happen to think Butler is an artifice coming from doing the calculations by hand, as it involves substantially less calculations. These days, computers can do that.

 

In short: there is no average to toss calculations from. Extreme (and every) results weight are controlled by the size of the sample, which I think was kept at 16 in non-tourney tables to keep the Movie fully contained in a single view (I think next good number would be the next power of 2, 32, which wouldn't enter in a single view, without scrolling).

 

Tourneys are managed different, the scores you get initially are from your own section (small sample), but just a guide, as the scoresare recalculated against all sections at the end.

 

* The Cavendish uses the same method, without dividing, so their numbers appear bigger (and very different than Butler). BBO divides, so numbers appear similar to Butler.

[matmat]

the ones I am talking about are the +2200 etc. scores when the whole field is in partials. seems crazy to be losing 3 imps on a board where all the results are partscores and someone bid up to 7n and claimed 2 or 3 tricks...

[Tola18]

the ones I am talking about are the +2200 etc. scores when the whole field is in partials. seems crazy to be losing 3 imps on a board where all the results are partscores and someone bid up to 7n and claimed 2 or 3 tricks...

Yes, or say your partner had been passed off a que-bid, slam going in the correct suit. There is no reason to play this. If opps dont accept to redeal you can as well immediately claim 0 tricks as cold bottom is cold bottom even at IMPs.

 

Or shall we account there are always some who pass on que-bids?? And therefore even these will usually get some points? (and lesson? Do always play on...Even in the most horrible imaginable contract).

 

 

Second. In Imps across the field in f2f bridge it IS usual to take away the two extreme-results, the best and the worst.

 

So this proposal do have some good merits.

[barmar]

the ones I am talking about are the +2200 etc. scores when the whole field is in partials. seems crazy to be losing 3 imps on a board where all the results are partscores and someone bid up to 7n and claimed 2 or 3 tricks...

In IMP Pairs scoring you're not really "losing" anything when you get negative scores. Remember, all that matters is your score relative to other pairs, and most of the field got that same -3 IMPs (except for the pair that was given the huge gift). If you ignore the +2200 you'll still all get about the same results, so the total IMPs won't change much.

 

If you throw out the extremes, what happens to the pairs at the +2200 table? Do they not get their +/-20 IMPs?

[Walddk]

If you throw out the extremes, what happens to the pairs at the +2200 table? Do they not get their +/-20 IMPs?

Sure they do; you just get a different datum (average) score. The 2200 will be compared to that score. The advantage is that those extreme results won't affect the "normal" results.

 

In a very strong field with up to 8 tables it should not be necessary to toss any results before you calculate the datum. We don't have a world class field on BBO, so in my view, playing IMPs, it makes a lot of sense to throw away the extremes, perhaps even 2 from both ends.

 

Roland

[winkle]

I think we should throw out the middle 14 scores because most of the time they're crazy.

[inquiry]

Let me give a hypothetical... suppose you bid to a grand slam on great fit with a great auction and only 24 hcp. No one else even bid slam, and some missed game. So you are looking at a fair and just +2210 verus something like 540 average otherwise.

 

How is this scored at imps? At matchpoints?

 

Clearly you deserve a 100% matchpoints. And if you get it, how was thowing it out affected?

 

And at imps? You are plus 2000 against those not in game, and plas 1500 versus the others. If you get your plus imps, how does this adjustment occur? You are still plus a big bundle.

 

I am against "throwing out" top and bottom. What I am for, is for "throwing out" (well, temporary suspension at least) people who intentionally distort the scores (opening 7NT on 5 count in first seat, for instance).

[barmar]

If you throw out the extremes, what happens to the pairs at the +2200 table?  Do they not get their +/-20 IMPs?

Sure they do; you just get a different datum (average) score. The 2200 will be compared to that score. The advantage is that those extreme results won't affect the "normal" results.

But since BBO computes cross-IMPs, not IMPs against a datum, this doesn't apply.

 

In fact, when using the datum method it IS popular to toss out the extremes when computing it. I played in a club IMP pairs game yesterday, and the director was still using the datum method (when I asked him, he'd never even heard of cross-IMPs, and didn't realize ACBLScore can do this), and he said he threw out the extremes (obviously he DID know how to make ACBLScore do this).

[Rain]

I don't believe in tossing out scores for the most part. What I believe in is

 

1) GIB player scores should be scored with the field, but not taken into account for the field. (tossing out score)

 

2) More comparisons to level out averages because of disparity in standard.

 

It seems like this is a recurring point brought up every 12 months. Is there something about the autumn air? =P

[PeterE]

Second. In Imps across the field  in f2f bridge it IS usual to take away the two extreme-results, the best and the worst.

No, it IS NOT :D At least not in IMPs across the field in Germany and I can't imagine it is elsewhere ...

Perhaps to show the difference between Butler (scoring against a datum) and IMPs-a-t-f (scoring directly against the other scores):

 

You have 7 scores: +2200 +300 +170 +170 +140 -50 -1400

 

In Butler you compute the datum, i.e. the arithmetic mean of 'all' scores. This can be done with regard to the extremes or without, so:

 

datum with extremes: 1530 : 7 = 218,6 rounded 220

datum without extremes: 730 : 5 = 146 rounded 150

 

Now the IMPs for NS will be (first with extremes then without in brackets)

 

+2200: 18 (19)

+300: 2 (4)

+170: -2 (1)

+140: -2 (0)

-50: -7 (-5)

-1400: -17 (-17)

 

So, in Butler (datum) scoring there is a difference (of course, because you are changing the imaginary score of your "teammates". The effect is, when you compute the datum without the extremes you get more "fair" scores in the "normal" score-range.

 

In IMPs-across-the-field you compare your own score with every other score "in the field", convert the difference in IMPs and either you take this result as the score or you divide it by the number of results (to get a score "like" a single comparison). As you compare your score with every other, you cannot skip the extreme scores !

 

+2200: vs +300 (18), vs +170 (19), vs +140 (19), vs -50 (20), vs -1400 (22)

==> 18+19+19+19+20+22 = 117 absolute and 16.7 normalized

+300: vs +2200 (-18), vs +170 (4), vs +140 (4), vs -50 (8), vs -1400 (17)

==> -18+4+4+4+8+17 = 19 absolute and 2,7 normalized

+170: -19-4+0+1+6+17 = 1 aboslute and 0,1 normalized

+140: -19-4-4+5+17 = -5 absolute and -0,7 normalized

-50: -20-8-6-6-5+16 = -29 absolute and -4,1 normalized

-1400: -23-17-17-17-17-16 = -107 absolute and -15,3 normalized

[1eyedjack]

Add my voice to NOT throwing out the extremes

 

Suppose there is a cold slam that only one pair bids, for +1431 to N/S, and the rest of the room making +680 to N/S. If you throw out the extremes then the datum is 0 to N/S. Is that fairer than getting the negative score that you deserve for not bidding it? The negative score that you actually achieve in cross IMPs would be quite small, as clearly the slam is a difficult one to bid, as is evidenced by the fact that only one bid it. So you get a negative score, which you deserve because you missed a slam, and that negative score is limited by the fact that you are in good company. All is fair.

 

Suppose that someone goes for 2200 in in ill-judged sacrifice. Why should you suffer any more or less than the players sitting at the table where it occurs? In all other areas the results at other tables have a direct impact on your own table's results. The Cross IMP scoring simply simulates the fact that you are taking part in 15 simultaneous mini-teams matches each of 2 tables.

 

There is also the distortion of the logarithmic scale to consider, as mentioned at:

http://www.thinkstop.com/bilbridge/boards/...770&postcount=3

[bid_em_up]

I think we should throw out the middle 14 scores because most of the time they're crazy.

LOL!

 

 

Sad but true....

[barmar]

Suppose there is a cold slam that only one pair bids, for +1431 to N/S, and the rest of the room making +680 to N/S. If you throw out the extremes then the datum is 0 to N/S. Is that fairer than getting the negative score that you deserve for not bidding it?

I think the results you get are more like team game scores when you throw out the extremes. In this example, the pair that bids the slam gets +13 IMPs, all the others get 0. That's the same IMP difference as they would get in a team game if they bid the slam at their table, but it wasn't bid at the other table. If you keep the extremes, the datum with 16 tables would be 730. The pair bidding the slam would get +12, and all the game bidders would get -2. This is a 14 IMP difference.

 

And in cross-IMPs, the slam bidders get +13, the non-slam bidders get -1.

[1eyedjack]

Suppose there is a cold slam that only one pair bids, for +1431 to N/S, and the rest of the room making +680 to N/S.  If you throw out the extremes then the datum is 0 to N/S.  Is that fairer than getting the negative score that you deserve for not bidding it?

I think the results you get are more like team game scores when you throw out the extremes. In this example, the pair that bids the slam gets +13 IMPs, all the others get 0. That's the same IMP difference as they would get in a team game if they bid the slam at their table, but it wasn't bid at the other table. If you keep the extremes, the datum with 16 tables would be 730. The pair bidding the slam would get +12, and all the game bidders would get -2. This is a 14 IMP difference.

 

And in cross-IMPs, the slam bidders get +13, the non-slam bidders get -1.

*IF* the proposal were to IMP against a datum (ie Butler style), *THEN* I would agree that tossing out the extremes (in order to determine the datum) would be sensible.

 

In cross-IMP scenario, before dividing the total score by the number of comparisons, the pair that score +1430 get a net +195 IMPs, being 15 distinct and separate matches each scoring +13 IMPs. Each other N/S gets -13 IMPS being 14 flat matches and one 13 IMP loss. There are 15 such scores of -13 IMPs, so the total IMP allocation is all square. It is sensible to divide by the number of comparisons simply so that no hand receives undue weighting by reason of differences in the number of times that hands are played, but that is really tangential to the main issue, which is whether it is intrinsically fair or unfair for the numerator to evaluate as above. At the moment I cannot see the objection.

[pclayton]

Look, just cross IMP over a bigger number of boards.

 

2200's don't happen on every board, but they happen often enough to skew normal results. When BBO had 20 tables in the MBC, 16 comparisons made sense. In a national, its not uncommon to be scored over 3, 4, 5 or even 6 sections (the latter being 78 pairs in your direction).

 

Uday, I hope that a bigger comparison list is at the TOP of the software updates.

[DrTodd13]

At MPs, there is no need to throw out extremes. A top board of 600 versus 2200 doesn't affect the rest of the field. At IMPs, with a small number of plays per board, the 2200 score can really skew things. In the long run it doesn't matter but there is always the day where the cards don't run your way and the opps bid to unbeatable game after unbeatable game and you wind up 50IMPs down for the day for the sole reason that a bunch of intermediates can't manage to bid game with 28 points or they screw up the play. You know if opps bid game and make it that 95% of the time you expect to lose 3 IMPs. You should punish the people who fail to bid game and not reward those who do the normal thing. Therefore, my opinion is that your IMP score should be based on the median result and not the mean result. If N tables have played the board, compute everybody's differential by comparing them against (N/2)th best score.