A multiplicative model for GOE?

[Mathman]

I am always a little bit bugged by attempts to add together disparate quantities. Factored ordinals were OK because the things that were combined were of the same kind. But ADDING the base value (what did you do) to the GOE (how well did you do it) seems like apples and oranges.

What would you all think of the scoring conventions used in diving? There is the difficulty score and then that is multiplied by the execution score.

In figure skating it would work like this. If you do a triple Lutz, you start with a base value (difficulty score) of 6.0. Then you get a grade of execution score of 100, 90, 80, … down to 0 for a failed attempt. 100 means perfect. You score 100% of the element’s base value. If it is excellent but not 100%, then you get 90. 90% of 6.0 is 5.4. That is your score for the element.

 

[blackey]

I can see what you mean, but it just way too complicated... I mean I'm that kind of nerd who would love to digging into those scores but I just don't think ISU wanna the system to be that hard to understand.

 

[4everchan]

I like both diving and gymnastics scoring way better than 6.0 or IJS.

I never understood why there wasn't a proper way to evaluate Figure Skating. COP is probably the lesser evil right now, but it doesn't mean there shouldn't be another system into place.

 

[Nathan13]

I am always a little bit bugged by attempts to add together disparate quantities. Factored ordinals were OK because the things that were combined were of the same kind. But ADDING the base value (what did you do) to the GOE (how well did you do it) seems like apples and oranges.

What would you all think of the scoring conventions used in diving? There is the difficulty score and then that is multiplied by the execution score.

In figure skating it would work like this. If you do a triple Lutz, you start with a base value (difficulty score) of 6.0. Then you get a grade of execution score of 100, 90, 80, … down to 0 for a failed attempt. 100 means perfect. You score 100% of the element’s base value. If it is excellent but not 100%, then you get 90. 90% of 6.0 is 5.4. That is your score for the element.

Interesting idea.

Would a 3Lz with, for example, arms over the head and a difficult entry be worth 100%, while a clean Lz with good height, etc. would be less, or would the base value change for difficult execution?

 

[Mathman]

Interesting idea.

Would a 3Lz with, for example, arms over the head and a difficult entry be worth 100%, while a clean Lz with good height, etc. would be less, or would the base value change for difficult execution?

Hm. Good question. My first instinct would be to say forget the arms over the head. That could be rewarded in presentation or musical interpretation if it contributes to the program.

 

[Mathman]

I never understood why there wasn't a proper way to evaluate Figure Skating. COP is probably the lesser evil right now,..

I have long since given up the fight, but ordinals (which served the sport for 100 years) do make sense. The purpose of having a scoring system at all is to determine who skated best. This is quality: best, second best, third best. Not quantity: 137.29 points 121.06 points, etc. Oh well.

 

[4everchan]

I have long since given up the fight, but ordinals (which served the sport for 100 years) do make sense. The purpose of having a scoring system at all is to determine who skated best. This is quality: best, second best, third best. Not quantity: 137.29 points 121.06 points, etc. Oh well.

what about a system like in moguls

3 scores make up the total.... and there is factoring for the degree of difficulty

jumps (2 jumps are performed and they have Degrees of difficulty IIRC)
speed (timed performance, so purely objective)... in figure skating there could be a few things measured objectively nowadays... speed and height of jumps could be evaluated numerically.
turns (these are evaluated by judges, and could equal skating skills)

like in every judged sport, people will always complained, but i hear it less often in other sports than in figure skating..
i know you liked ordinals but with the easy corruption it offered, ISU will never go back to a similar system..

Finally. and for what it's worth, I used to watch figure skating with my landlady (RIP) and she understood nothing in the scoring now in the IJS.. but she was following the scores, adding the SP + LP and was completely fine with 187.02.... or 174.90... it was clear for her, that one was higher than the other... and much easier to deal with than ordinals... she was in her 85-90s when we watched skating together and told me that she didn't understand a single thing about Calgary ordinals and points but now could make it happen with the "new system"

 

[andromache]

I have long since given up the fight, but ordinals (which served the sport for 100 years) do make sense. The purpose of having a scoring system at all is to determine who skated best. This is quality: best, second best, third best. Not quantity: 137.29 points 121.06 points, etc. Oh well.

PCS scores are way more problematic in this regard than GOE. GOE at least have criteria that are semi-objective.

 

[NanaPat]

Currently it looks like the GOE and BV are added, but that's not really the case. The "pure GOE" (-3 ... +3) is first multiplied by a factor (which is related to the base value) before it is added. You could probably come up with the same result by first changing the GOE into a multiplier instead of an "adder". see what I mean. It isn't a true sum of the GOE and BV because the GOE is altered first.

For example (and these are made up, for discussion only, and not accurate): an element of BV 2.0 and GOE -3 may wind up worth 1.25 (because the -3 is "discounted" to -.75 before it is added) but an element of BV 4.0 and GOE -3 end up worth 2.5 (the -3 is discounted by a smaller amount to -1.5) and an element of BV 8.0 and GOE -3 end up worth 5 (the -3 isn't discounted at all in this case). You're not just subtracting 3 in every case, but are subtracting more if the BV is bigger.

 

[moriel]

I am always a little bit bugged by attempts to add together disparate quantities. Factored ordinals were OK because the things that were combined were of the same kind. But ADDING the base value (what did you do) to the GOE (how well did you do it) seems like apples and oranges.

What would you all think of the scoring conventions used in diving? There is the difficulty score and then that is multiplied by the execution score.

In figure skating it would work like this. If you do a triple Lutz, you start with a base value (difficulty score) of 6.0. Then you get a grade of execution score of 100, 90, 80, … down to 0 for a failed attempt. 100 means perfect. You score 100% of the element’s base value. If it is excellent but not 100%, then you get 90. 90% of 6.0 is 5.4. That is your score for the element.

I think first step would actually make the GOE calculation automatic, leaving to judges to tick boxes.
And see how this works.

Because currently, the issue is not how GOE is calculated and added, but that it is not always clear why a certain value was given in first place. And we are left to wonder why a jump received all +3 across the board.

 

[AnalyticalMind]

I mean, GOEs are basically multiplicative bonuses (with bad rounding) right? You're giving a bonus over base value that's adjusted (somewhat incompletely) according to the base value. A +3 on a non-axel triple gets you 2.1 points over a BV of 4.1-6.0 (so a bonus of 35~51%), whereas a +3 on a non-axel quad gets you 3.0 over a BV of 10.3-13.6 (so a bonus of 22~29%). GOEs would be perfectly multiplicative if, for instance a +3 got you +30% over BV regardless of the element. So, it's the "progressive" choice of values that's a bit off, but there's nothing in the system itself that's a priori mathematically flawed, as you suggest.

An interesting question to be asking, then, is why they don't go for a purely multiplicative GOE (e.g., the +3 GOE = +30% BV regardless of element).

So it's interesting to consider what effect this has. Note that a +3 3Lz could get you 8.1 points, which is close to the BV of a 3A (8.5). As indicated by this and the examples given above, this system gives more of a bonus for a well-executed triple than a well-executed 3A or quad, and I think that could be precisely the intent. By making the GOE bonus that you can get on a lower-BV element worth more than the bonus that you can get on a high-BV element, they're effectively building into the system a mechanism to allow for the Jason Browns of the world to have a fighting chance over a quad-jumper with little by way of +GOEs (i.e., the gap in TES wouldn't be insurmountable by superior PCS).

===
Source for BVs / GOEs: ISU Communication No. 2000 (http://static.isu.org/media/1003/2000-sptc-sov-and-goe-2016-2017_revised-july-14.pdf)

 

[Sackie]

No matter what scoring system is put in place the judges will always find a way around it. They have always found a way for their favs to win medals when possible - Browning won a couple he shouldn't have, as did Kwan. And Hughes would never have won an Olympic title had she been marked correctly in the short program. One thing that the COP has done however is make it harder for them to hold a skater up too much. Under the old 6.O system they had to make deductions for missed elements in the short program but that simply meant they would mark others down (even when they skater clean programs) to keep their favs close to the top when they had an error. Now it is not as easy to hold a skater down when they skate clean. Daleman I think is proof of this. I really believe if we were still under the 6.0 system she would not have been 3rd in the short program.

 

[4everchan]

One thing that the COP has done however is make it harder for them to hold a skater up too much. Under the old 6.O system they had to make deductions for missed elements in the short program but that simply meant they would mark others down (even when they skater clean programs) to keep their favs close to the top when they had an error. Now it is not as easy to hold a skater down when they skate clean. Daleman I think is proof of this. I really believe if we were still under the 6.0 system she would not have been 3rd in the short program.

pretty much agree with this... ice dance is so much more exciting with COP as well... with 6.0, Gabby and Guillaume would be about 5th or 6th by now... gotta wait your turn

 

[Weathergal]

pretty much agree with this... ice dance is so much more exciting with COP as well... with 6.0, Gabby and Guillaume would be about 5th or 6th by now... gotta wait your turn

For me, this is one of the big pluses of CoP for me - it has made ice dance so much more interesting and has made things like Gabby and Guillaume's meteoric rise possible. I know CoP has issues, and I think it desperately needs to be refined to address those issues, but this is really a positive.

As a side note, a few of the things that I believe should be addressed: runaway PCS and PCS being used to reward things it wasn't intended to, cookie cutter programs, transitions just thrown in to add to the total but not done with any meaning (ex. spread eagles, ina bauers, and spirals with little extension and ones that are barely held any length of time--pet peeve!), backloaded programs (I think the idea was to avoid front-loaded programs but now instead of balanced ones, we're evolving to backloaded ones. While I appreciate the mental strength and conditioning required to pull that off, isn't that really trading one issue for another?).

 

[Sackie]

As a side note, a few of the things that I believe should be addressed: runaway PCS and PCS being used to reward things it wasn't intended to, cookie cutter programs, transitions just thrown in to add to the total but not done with any meaning (QUOTE]

The one thing that I have noticed with any judging in Skating is that it moves slowly. When the men started doing 3As they were the ones all of a sudden that were jumping to the top of the pile even with bad programs. The judges were willing to over look poor transitions and footwork with the idea that the skater needs to leave these out in order to land a clean 3A. When the 1st Quads came in they did the same thing and now they are doing it with programs with multiple quads. But once more and more skaters are adding multiple quads the judges will begin to look at the other stuff again to balance things out. After all Judges are human too and they too must be a little awed by skaters doing 3 or 4 or even 5 quads in a program. I mean some of them were in the system when they were only doing basic triple jumps with out even 3/3 combos.

 

[Weathergal]

As a side note, a few of the things that I believe should be addressed: runaway PCS and PCS being used to reward things it wasn't intended to, cookie cutter programs, transitions just thrown in to add to the total but not done with any meaning (QUOTE]

The one thing that I have noticed with any judging in Skating is that it moves slowly. When the men started doing 3As they were the ones all of a sudden that were jumping to the top of the pile even with bad programs. The judges were willing to over look poor transitions and footwork with the idea that the skater needs to leave these out in order to land a clean 3A. When the 1st Quads came in they did the same thing and now they are doing it with programs with multiple quads. But once more and more skaters are adding multiple quads the judges will begin to look at the other stuff again to balance things out. After all Judges are human too and they too must be a little awed by skaters doing 3 or 4 or even 5 quads in a program. I mean some of them were in the system when they were only doing basic triple jumps with out even 3/3 combos.

Interesting thoughts - I hope you're right! :agree:

 

[Mathman]

Currently it looks like the GOE and BV are added, but that's not really the case. The "pure GOE" (-3 ... +3) is first multiplied by a factor (which is related to the base value) before it is added...

I mean, GOEs are basically multiplicative bonuses (with bad rounding) right?...

I am sure that I am the only person in the skating world who is bugged by this issue. But here's the thing. A weighted sum is still a sum. The weighting factor carries no units.

Suppose you want to add 3 inches to 4 pounds. It doesn't help to first reduce the 4 pounds by a factor of 80% first, then add 3 inches to 3.2 pounds. You still face the same problem. Now you have a "sum" of 6.2 ... whats?

The suggestion is to turn the GOEs into the (unit free) weights themselves. So if you do a triple Lutz with GOE of 100% and a triple Salchow with a GOE of 60% and a combination spin with a GOE of 80%, you end up with

(base value of 3Lz)x1 + (base value of 3S)x.6 + (base value of combo spin)x.8 -- and the answer is so many whats? Technical points.

The idea is that GOE is part of technique, just the same as correct edge, full rotation, etc.

 

[Mathman]

I can see what you mean, but it just way too complicated... I mean I'm that kind of nerd who would love to digging into those scores but I just don't think ISU wanna the system to be that hard to understand.

I don't think it is hard to understand at all. The skater tried a triple Lutz but he had to save a wobbly landing. He gets 60% of the full base value.

Now he does a very good triple Salchow. He gets 90% of full base value.

Now he tries a triple flip, but he under-rotates badly and falls. This is a failed element. He gets 0% of full base value.

 

[gkelly]

The base values are measures of difficulty.

The GOEs are primarily measures of quality, of different kinds, both errors (of varying severity) and excellence or enhancements.

Base value = executed successfully according to the requirements of that kind of element. I don't see any reason to change that.

We don't start with the best possible example of the element for the base value, because almost no one would be able to earn full value that way. And someday someone will find a way to do that element even better.

I definitely think that better-than-acceptable qualities of individual elements should be rewarded in the score for those individual elements and not in the PCS. What if the skater does a very good first element, falls hard on the second, and then struggles with the rest of the program after that? PCS will probably be significantly lower than for a clean performance -- but that first element was just as good as in the clean performance and it's only fair to the skater to recognize and reward that even while penalizing for the problems later on.

Often the same elements has both good/very good/excellent qualities and also some weaker qualities or punishable errors. So how should the system allow for both rewarding and penalizing different aspects of the same element?

It would make sense to me to use percentages of base value to reflect reductions for errors. That is already the case for underrotations and blatant wrong edges. Other errors could each have a percentage reduction, and some of them have a range of reduction based on severity, which is a judgment call that judges might not agree on.

The simplest way to adopt that would be to continue to have the tech panel and judges do what they're already doing, but instead of listing the actual values of the reductions for each element in the Scale of Values, just write at the top of the SoV that, for example, each negative GOE step always takes away 25% of the base value and each positive GOE step always adds 25%, and then do that factoring either to each judge's GOE+base value before averaging or average the GOEs and then factor the panel's average GOE by percentage rather than by the rough proportional values now listed in the SoV. Or determine that a flat 25% per GOE step is not appropriate but some kinds of elements should have higher or lower percentages of penalty or reward based on quality. (For example, the Choreo Sequence, where everyone gets the same base value and the GOE is intended to be the source of the distinctions.)

Asking judges to tick boxes for every positive and negative quality they want to reward or penalize and assigning mathematical ways to adjust the scores accordingly would be a theoretically sound approach and would certainly give everyone a lot of information about the reasons for the scores. But I don't see how it would be practical in real time.

 

[Mathman]

The base values are measures of difficulty.

The GOEs are primarily measures of quality, of different kinds, both errors (of varying severity) and excellence or enhancements.

Base value = executed successfully according to the requirements of that kind of element. I don't see any reason to change that.

To me, the reason why the ISU might want to change that is that they might feel that a skater does not deserve the full base value for just somehow getting through the minimum definition of the element. Or maybe I would say it this way: Things like good height and distance, good flow on the landing, etc., should be part of the definition of "difficulty" that the base value is based on. It is more difficult to do a good triple Lutz than to do a bad one. Why couldn't this be taken into account when awarding a base value based on the difficulty of the element?

As for embellishments like a 'Tano, I am not really impressed by making something more difficult for no other reason than to make it more difficult. You could skate with one arm tied behind your back, but I would not be in favor of giving extra points to a skater who does so.

We don't start with the best possible example of the element for the base value, because almost no one would be able to earn full value that way. And someday someone will find a way to do that element even better.

I do not see any problem with either of these. If only a few skaters can do a 100% triple Lutz and the rest can do only a 90 percenter, then good for the champion and it gives the runner-up something to work on.

I don't think that someone in the future will do a triple Lutz that is better than any triple Lutz that has ever been done in the history of skating. A quad Lutz, maybe. But the judges should be judging the quad Lutzes that come before them with a mental picture of what a "perfect" one should look like.