[WellSpyder]

mjj29, on 2012-February-13, 03:44, said:

Surely poor players don't worry about the scoring at all though and would be happy with any method of scoring. At best they want to know 'is it IMPs or pairs', although most of them don't change their play either way...

Indeed. We play Butler pairs once a month at my club, but the Bridgemates still display match point percentages when scores are entered. I'm quite struck by how many people take a serious interest in whether they have scored, say, 50% or 60% on a hand, even though this has virtually nothing to do with their imp score on the hand.

[barmar]

mjj29, on 2012-February-13, 03:44, said:

Really? I don't understand the concept of a datum - I have no idea why you would consider the mean of the available scores a reasonable thing to imp against or why it's reasonable to discard that number of outliers. I don't understand why it's ever reasonable to imp against a score that noone (including the computer) actually got on that board.

In Butler scoring, you treat "the field" as the opposing team. So to get their score, you average all the results. The metaphor would work better if you excluded your own score when computing the datum to IMP against, but that would also negate much of the calculation benefit of using the datum.

But as I mentioned earlier, I think the real reason for Butler scoring was because it was developed before computer scoring. Calculating cross-IMPs takes N times as along as Butlers, where N is the number of tables. But once computers took over the job of scoring, it's mainly inertia that keeps it around. Most players don't really think about the mathematics of it, they just like what they're used to.

[mycroft]

I would have thought that teams-of-8 were, actually, "score all 4 matches and combine/average the IMPs" - computers should be able to do that straight up, right?

I don't see a problem with any scoring method, frankly, but the IMP table *is* designed for head-to-head results, rather than any combination, be it datum or "add the scores and IMP the difference". It doesn't matter, however, as long as it's known in advance, and people can play to it.

[Vampyr]

WellSpyder, on 2012-February-13, 03:31, said:

FWIW, I feel strongly that Cross-imps are better than Butler imps, and suggest changing from Butlers whenever the opportunity arises. But there is usually very little enthusiasm for this since everyone is used to Butlers and most don't care about the distortions involved.

Cross-imping is now the norm in my part of the world, which is not too far from your part of the world.

[Trinidad]

barmar, on 2012-February-13, 10:59, said:

In Butler scoring, you treat "the field" as the opposing team. So to get their score, you average all the results. The metaphor would work better if you excluded your own score when computing the datum to IMP against, but that would also negate much of the calculation benefit of using the datum.

But as I mentioned earlier, I think the real reason for Butler scoring was because it was developed before computer scoring. Calculating cross-IMPs takes N times as along as Butlers, where N is the number of tables. But once computers took over the job of scoring, it's mainly inertia that keeps it around. Most players don't really think about the mathematics of it, they just like what they're used to.

The point of Butler is that you do not treat the field as the opposing team. Instead, you create some weird average as "the opposing team".

Would anybody get the idea of scoring "Board a Match Pairs"? (All pairs are scored against the datum, scoring 1 for a win, 0 for a loss and 1/2 for a tie.) Of course not. It is obvious that you get a better way of seeing who is best if you compare against the whole field, rather than against some average. For IMPs, this is not different. You should use the whole field as the opposing team, instead of only the "average" of the field. And then you get to Cross-IMPs.

Rik

[bluejak]

mjj29, on 2012-February-13, 03:44, said:

Really? I don't understand the concept of a datum - I have no idea why you would consider the mean of the available scores a reasonable thing to imp against or why it's reasonable to discard that number of outliers. I don't understand why it's ever reasonable to imp against a score that noone (including the computer) actually got on that board.

That is because you are not typical. Poor players find imps much much easier to understand where it is their score imped against another score: the datum.

mjj29, on 2012-February-13, 03:44, said:

Cross-imps, on the other hand, are both very easy to explain to people ('you imp against everyone and take the average') and seem much more sensible when you consider them in any detail.

Far more difficult, far less comprehensible. Not to you, but I am not talking about you.

mjj29, on 2012-February-13, 03:44, said:

Surely poor players don't worry about the scoring at all though and would be happy with any method of scoring. At best they want to know 'is it IMPs or pairs', although most of them don't change their play either way, it just gives them a different set of excuses for bidding failing games. Why, therefore, should we give them a statistically bogus method of scoring, rather than a reasonable one?

First of all, it is not statistically bogus, so stop trying to sneak such comments in. It is statistically fine. What you mean is that you and good players believe cross-imps is better, and can produce a justification.

Secondly, to repeat, poor players understand imps as a method of taking their score, another score, and there you are: imps.

[barmar]

Trinidad, on 2012-February-14, 06:15, said:

The point of Butler is that you do not treat the field as the opposing team. Instead, you create some weird average as "the opposing team".

I don't understand this response. If you want to treat the field as the opposing team, and IMP against it, you need a single score. How do you turn a dozen scores into a single score? Averaging is something most people think they understand, it's used in all walks of life. It's not always appropriate (have you ever met a family with 2.4 kids?), but it's ubiquitous.

[mycroft]

Well, except for the fact that those 2.4 kids are based on all families except the 100 000 with the most and the 100 000 with the least number of kids. Which might, just, skew the number a little.

I've found that Butler makes sense to the people who have had Butlered IMPs explained to them when they first played it, and have played it ever since; and cross-IMPs make sense to the people who ask me 15 minutes to gametime, the session after, "why did I get scores like -1.85 and 3.62?" Usually if one of those asks the long-timers, they get the explanation of how Butler works; if I hear that discussion, I try to move in to correct it, because "we don't do that any more".

Fortunately, that's becoming less common as we haven't had a Butlered IMP pairs game in this area for 8 or 9 years.

[campboy]

bluejak, on 2012-February-14, 09:44, said:

First of all, it is not statistically bogus, so stop trying to sneak such comments in. It is statistically fine. What you mean is that you and good players believe cross-imps is better, and can produce a justification.

I do not claim to be a "good player", but I am a good mathematician. What we mean by "statistically bogus" is Butlering has very poor statistical properties, such as imbalance of NS and EW scores, high sensitivity to changes in distant scores, distortion of the IMP scale and discretisation effects. What is your justification for saying it is "statistically fine"?

[bluejak]

Same for everyone.

[PeterAlan]

bluejak, on 2012-February-14, 19:24, said:

Same for everyone.

So's the same score on every board for everyone, N! points for the winner (however defined) of Board N, or any other idiotic scoring system one could dream up that applied equally to all participants. I think you need a better reason.

[c_corgi]

If, as seems both reasonable and inevitable, we persist with using IMPs for anything other than long head to head matches, shouldn't we just accept that the results will be less "pure"?

[Trinidad]

c_corgi, on 2012-February-14, 20:25, said:

If, as seems both reasonable and inevitable, we persist with using IMPs for anything other than long head to head matches, shouldn't we just accept that the results will be less "pure"?

Cross-IMPs are certainly not less "pure" than head to head team of four IMP scoring. As a matter of fact, Cross-IMPs are much "purer" than head to head teams of four because the data set is much larger.

So, there is no argument against using IMP scoring for pairs competitions. The only point is that you should use the mathematically correct method to do it. Cross-IMPs are mathematically correct and Butler is mathematically awful.

Rik

[bluejak]

PeterAlan, on 2012-February-14, 19:59, said:

So's the same score on every board for everyone, N! points for the winner (however defined) of Board N, or any other idiotic scoring system one could dream up that applied equally to all participants. I think you need a better reason.

I have already given perfectly good reasons why customer satisfaction is important in such decisions, rather than mathematical irrelevancies. Comparing with something people will hate hardly seems relevant. If that is your best argument, which need I bother?

[pran]

bluejak, on 2012-February-15, 08:53, said:

I have already given perfectly good reasons why customer satisfaction is important in such decisions, rather than mathematical irrelevancies. Comparing with something people will hate hardly seems relevant. If that is your best argument, which need I bother?

I would expect (assume) that those who "hate" Cross IMPs do so because of ignorance and "fear of the unknown". This is no argument against a "better" scoring method.

Count me in with those who prefer Cross IMPs for Butler, except that I favour "Normalized" Cross IMPs where the total amount of IMPs on a board is divided by the number of comparisons on that board for easier recognition of the results.

[bluejak]

We are not discussing personal preference. Of course I prefer normalised cross-imps. But I believe lesser players prefer Butler and that is the vital argument: they are the most important customers.

[Zelandakh]

bluejak, on 2012-February-15, 09:24, said:

We are not discussing personal preference. Of course I prefer normalised cross-imps. But I believe lesser players prefer Butler and that is the vital argument: they are the most important customers.

When I started playing bridge the same players had a strong preference for rubber bridge. That has not stopped many developing an appetite for duplicate in the intervening years. The same is likely to happen for Butler vs cross IMPs if clubs start offering the choice. The game of bridge does not have to stand still just because it has an older player base!

[pran]

bluejak, on 2012-February-15, 09:24, said:

We are not discussing personal preference. Of course I prefer normalised cross-imps. But I believe lesser players prefer Butler and that is the vital argument: they are the most important customers.

I was just wondering:

When you assign a weighted adjusted score at IMPs do you then use the weighted average of all appliccable total points result as one datum and assign the IMP result against this datum, or do you calculate the IMP results against each applicccable total points result and assign the weighted IMP average result?

I assume you agree that the first alternative corresponds to Butler scoring while the second corresponds to (normalized) IMP across the field, and that it is not insignificant which alternative is used?

Is the alternative to be used specified in EBU regulations?

[Vampyr]

Trinidad, on 2012-February-15, 05:10, said:

Cross-IMPs are certainly not less "pure" than head to head team of four IMP scoring. As a matter of fact, Cross-IMPs are much "purer" than head to head teams of four because the data set is much larger.

So, there is no argument against using IMP scoring for pairs competitions. The only point is that you should use the mathematically correct method to do it. Cross-IMPs are mathematically correct and Butler is mathematically awful.

Our local IMP pairs game uses "mathematically correct" cross-IMPs and then spoils it completely by arrow-switching.

[PeterAlan]

bluejak, on 2012-February-15, 08:53, said:

I have already given perfectly good reasons why customer satisfaction is important in such decisions, rather than mathematical irrelevancies. Comparing with something people will hate hardly seems relevant. If that is your best argument, which need I bother?

My point, bluejak, which seems to have totally passed you by, is that your "same for everyone" is totally inadequate as a criterion [for judging what is "statistically fine"]. Read what I said! I said nothing about anything else.

Whilst it's obviously necessary, it's very far from being sufficient.