Computer Scoring

[zasanya]

Setting

IMP Pairs 15 tables Skip Mitchell 15 EW phantom pair 12 rounds 24 boards

In the scoring program i use the computer gives following options among other things.

1) cross imp / comparisons

2) cross imp pairs / SQR(rc/2)

3) Butler Pairs

4) Aggregate scoring

Rightly or wrongly I chose method 2 although I do not know how it works.

On a particular board 3 pairs bid Vul game 8 pairs bid vulnerable slam and 1 pair bid grandslam.

The pair that bid grandslam was awarded 19.08 IMPS.

How is that possible. The datum is just about 1400.

Does method 2 have some different way of calculating datum? Should I use some other option?

Can someone please explain

[aguahombre]

Cross IMPs options involve a comparison of each pair's result against each other pair's result (with or without some kind of factoring thereafter). A datum comparison is different.

[Bbradley62]

Presumably, all 12 declarers at least made their contracts. What scores did those in game and those in small slam get?

[zasanya]

Presumably, all 12 declarers at least made their contracts. What scores did those in game and those in small slam get?

yes all made their contracts. The 12 scores were 1370/1390/720/1470/1390/1470/2220/1470/1440/620/640/1390.

The contracts were 3NT/5C/6C/6NT/7NT making 12 or 13 tricks (depending on whether Club Queen was guessed or not.

[Bbradley62]

Sorry, my question was: what IMP scores did the game bidders and small slam bidders get?

[mycroft]

So, if the word "datum" ever comes to mind, ignore it. Irrelevant to X-IMPs, bad idea anyway.

 

vulnerable, 7NT= will make 13 imps against all small slams; 17 against the games. Add 'em up, we get 8 13s and 4 3 17s, for 172 155. To get 19.08 IMPs, we'll be dividing by 9 8ish. I have no idea where that number comes from*, but it's sqrt(rc/2), and that is supposed to "reflect teams scoring". It looks like dividing by 9 8ish rather than 11 10 (EBU) or 12 11 (ACBL) is going to inflate the IMP score, but the key number isn't the 19.08, it's the 172 155.

 

[Edit: can't count to 12 after all these years of counting to 13. Argh.]

 

* GordonTD has it below. "rc" is 11*12. I have no idea why yet, but I'm sure all will be explained in time.

[barmar]

The datum is only used in Butler scoring. Cross-IMPs means that you compare a pair's score with all the other pairs playing the board, calculate an IMP difference for each of them, and add these all up to get a total. In methods 1 and 2 you then divide this total by some number to get an average IMP difference; in method 3 you just use the total.

[shyams]

So, if the word "datum" ever comes to mind, ignore it. Irrelevant to X-IMPs, bad idea anyway.

 

vulnerable, 7NT= will make 13 imps against all small slams; 17 against the games. Add 'em up, we get 8 13s and 4 17s, for 172. To get 19.08 IMPs, we'll be dividing by 9ish. I have no idea where that number comes from, but it's sqrt(rc/2), and that is supposed to "reflect teams scoring". It looks like dividing by 9ish rather than 11 (EBU) or 12 (ACBL) is going to inflate the IMP score, but the key number isn't the 19.08, it's the 172.

 

The method should be as Mycroft said. The actual x-IMP for the board should be roughly 14.1 (8x13+3x17)/11. It's not clear how the software showed 19.08.

Guess the correct option in the software for IMP scoring should be #1 - cross-imp comparisons.

[gordontd]

15 tables Skip Mitchell 15 EW phantom pair 12 rounds 24 boards

Really? How does that work?

[gordontd]

The method should be as Mycroft said. The actual x-IMP for the board should be roughly 14.1 (8x13+3x17)/11. It's not clear how the software showed 19.08.

(8x13+3x17)/sqrt((11x12)/2)=19.08

 

Guess the correct option in the software for IMP scoring should be #1 - cross-imp comparisons.

Yes :)

[mycroft]

Okay, thank you. Please explain where that (rc/2)^0.5 comes from, Gordon? And how rc = 11*12?

 

I found the manuals for this scorer, and it was - unhelpful on this point. Almost as if if we wanted to use this scoring method, we'd already know all about it. Sort of like a "multiple teams movement".

 

Again, though, it doesn't matter what the "divided by" is, at least for the scoring (okay, factoring, yeah); it's the raw IMP score from all the comparisons that matters. Bringing it down to a "reasonable" number that we're all used to from single-comparison teams scoring, by dividing by some reasonable, constant factor is just for our feeble brains.

[gordontd]

Okay, thank you. Please explain where that (rc/2)^0.5 comes from, Gordon? And how rc = 11*12?

r=results

c=comparisons

 

So I should really have written 12*11, not 11*12.

 

There has long been friendly disagreement between scoring experts about whether one should divide by the number of results or the number of comparisons, but the truth, as you say below, is that it doesn't matter much unless the numbers are small and you try to compare results scored by the two different methods.

 

I found the manuals for this scorer, and it was - unhelpful on this point. Almost as if if we wanted to use this scoring method, we'd already know all about it. Sort of like a "multiple teams movement".

 

Well since the scorer comes from England where "multiple teams movement" is standard terminology, it's hardly surprising that there's no need to explain it. However I think you are correct that no-one who didn't already know about this obscure scoring method would want to use it. The only document that I could find that mentions this formula does so in the context of a more complex discussion about constructing VP tables and doesn't really answer your question.

 

Again, though, it doesn't matter what the "divided by" is, at least for the scoring (okay, factoring, yeah); it's the raw IMP score from all the comparisons that matters. Bringing it down to a "reasonable" number that we're all used to from single-comparison teams scoring, by dividing by some reasonable, constant factor is just for our feeble brains.

[mgoetze]

So if you can't agree whether to use results or comparisons, you just take the geometric mean? Sounds like a fair solution.

[gordontd]

So if you can't agree whether to use results or comparisons, you just take the geometric mean? Sounds like a fair solution.

Except that they are dividing the geometric mean by sqrt(2), so the final results are all much larger than when dividing by either results or comparisons.

 

I'll try to find out more about this.

[zasanya]

Really? How does that work?

Sorry standard mithchell amd there was no phantom pair but that doesnt essentially change the IMP problem?

[zasanya]

Except that they are dividing the geometric mean by sqrt(2), so the final results are all much larger than when dividing by either results or comparisons.

 

I'll try to find out more about this.

Thank you Sir.

[gordontd]

Sorry standard mithchell amd there was no phantom pair but that doesnt essentially change the IMP problem?

No, it doesn't change it - I was just intrigued to find out more about this unknown movement!

[RMB1]

Really? How does that work?

I suspect 30 boards in play, but pairs play only 24.

[Vampyr]

I suspect 30 boards in play, but pairs play only 24.

 

For IMP pairs this is even worse than arrow-switching.

[aguahombre]

For IMP pairs this is even worse than arrow-switching.

Yes. And, what fun at cross-IMPs. There could be hundreds of IMPs available to some pairs and not to others.

[gordontd]

Except that they are dividing the geometric mean by sqrt(2), so the final results are all much larger than when dividing by either results or comparisons.

 

I'll try to find out more about this.

I have found out more. It was invented by Max Bavin for use when scores are converted to Victory Points to allow for the fact that a Victory Point scale should be affected by the size of the field as well as by the number of boards. Obviously when teams matches are Victory Pointed the field is always the same size, so this can be ignored, but for IMP pairs that is not the case. Using this divisor should mean that one VP scale can be used for each number of boards in a match, regardless of how big the field.

[Vampyr]

Yes. And, what fun at cross-IMPs. There could be hundreds of IMPs available to some pairs and not to others.

Quite. Seems like if you want that amount if randomness, you would have more fun trying your luck at the local rubber bridge club.

[barmar]

A friend of mine writes a bridge blog, and coincidentally his item yesterday was about the random swings that are likely in IMP Pairs at clubs.

 

http://jeff.bridgeblogging.com/2014/04/29/imp-pairs-event-at-the-club/

 

This is probably one of the reasons that clubs don't run IMP Pairs very often. Our club, with one weekly game of 8-10 tables, holds 3 of them a year (along with 1 individual, 2 team games, and 1 pro-am).