Ice scope data shows zero correlation between height & GOE

[Shanshani]

I don't see why the size of the skater should affect what is objectively considered a big jump--in sports like long jump, they don't score you differently based on how tall you are. In fact, I think that that actually speaks to a flaw in letting judges determine this subjectively--a jump may *look* higher because of the size of the skater, but not actually be higher.

I also think that, while yes, the correlation between GOE and size of jump should not be exceptionally high because they are not the only components of a jump's quality, the fact that there is virtually no correlation at all is concerning. Again, you are much better off guessing GOE based off PCS and BV than you are based on ice scope jump statistics--like, absurdly so. The relationship between PCS and GOE, and the relationship between BV and GOE, is quite strong, whereas the relationship between jump size and GOE barely exists. Right now, jump size explains somewhere in the neighborhood of 1% of the variation in GOE, 0% if we are considering height. That seems incredibly low.

In fact, let me re-iterate that if you remove Yuzuru Hanyu from the data set because he's a large outlier, there is actually a negative relationship between jump height and GOE. Essentially, if your name is not Yuzuru Hanyu, jumping higher is correlated with lower scores.

Yes, of course this analysis depends on the accuracy of the ice scope data. Having seen multiple measurements of the same skater jumping the same jump on different, the numbers seem fairly consistent and do not vary wildly, which seems to me to provide at least some evidence of its reliability. Additionally, the numbers seem quite plausible--we see, for instance, that Shoma doesn't jump very high, but he gets good distance and has very high speed exiting the jump, which matches the data given. Or Kolyada jumping high but not far, Yuzuru's 3A being gigantic, and so on. I'm certainly more inclined to trust it than fan made estimates using body-lengths or board-comparisons, which depend a lot on the perspective the shots are taken and are prone to being colored by fan bias.

 

[reneerose]

Although I find the axel cam/ice scope data entertaining, the noise that comes when it first appeared on my screen was quite distracting... :noshake:

 

[OniBan]

Is this true? There are some tiny ladies with big jumps (Liza - all of her jumps are huge and she is very short), and Shoma is also one of the shortest men and has a high 3A. Midori Ito as well. Alena Kostornaia is much shorter than Trusova and Shcherbakova, but her jumps are not small at all.

Part of sport is physical ability. Some people have better physical ability than others.

Actually, Shoma's 3A is not very high, bu it goes long across the ice, and exits with good speed.

I think it's really a combination of a lot of factors can attribute to differences in jumps - body types, physical abilities, technique, etc.
It's not just limited to one. Some people do have better physical ability than others, it is also possible that some part of this ability is affected by physical attributes.

 

[shyne]

Your model does not capture the correlation between landing speed, height and distance. It doesn’t make sense to plot in a simple linear regression

 

[Shanshani]

Your model does not capture the correlation between landing speed, height and distance. It doesn’t make sense to plot in a simple linear regression

What would you recommend then? The point of this post was to do a quick-and-dirty analysis while the data was fresh and on people's minds, but I'd welcome suggestions for a more robust analysis. But do note that multicollinearity (correlation between explanatory variables) is precisely why there might be problems with a multivariate analysis.

 

[Sugar Coated]

I’d be curious to know if there is any data on rotation speed and GOE? I’m wondering if judges are actually using this data when judging. I’d imagine that’s perceptually easier and more intuitive for someone to distinguish differences very quickly rather than something like height or distance of a jump which seems to require technological assistance.

Also, it’s been a long time since I’ve taken physics but doesn’t someone have to sacrifice height for distance and vice versa? I’d think you’d need to take both data points into account. Someone jumping high but not far or far but not high may not be receiving scores as high as someone who is utilizing both height and distance.

 

[elysium]

What would you recommend then? The point of this post was to do a quick-and-dirty analysis while the data was fresh and on people's, but I'd welcome suggestions for a more robust analysis. But do note that multicollinearity (correlation between explanatory variables) is precisely why there might be problems with a multivariate analysis.

Two immediate and simple extensions of the analysis could be:

1. Do a multidimensional linear regression with all 3 input parameters at once and then look at remainder, whether it fluctuates wildly.

2. Percentile tank all three of the inputs individually and then equal weight to get a total score. And check whether this score has any resemblance with the actual scoring order from the judges.

Likely both might show that the judges make random decisions.

Also they should measure the airtime during the jump, that would be a rudimentary proxy for the rotation speed. Although I am not sure personally that it should be counted as important for the judges, since in the end number of completed revolutions is what counts.

 

[Shanshani]

Two immediate and simple extensions of the analysis could be:

1. Do a multidimensional linear regression with all 3 input parameters at once and then look at remainder, whether it fluctuates wildly.

2. Percentile tank all three of the inputs individually and then equal weight to get a total score. And check whether this score has any resemblance with the actual scoring order from the judges.

Likely both might show that the judges make random decisions.

I'm going to have to familiarize myself with more complicated regressions more, but it turns out that height, distance, and landing speed aren't all that correlated, so multicollinearity isn't a significant problem. But the multiple linear regression I ran suggested the same thing as the initial analysis--none of the 3 variables are significant predictors of GOE. In fact, the calculator I ran just refused to include height and distance in the model at all due to the lack of effect of those two quantities on GOE I think it went for landing speed just because out of the three the correlation between it and GOE is the strongest by a hair. So the model just reduces down to the single-variable model for landing speed vs GOE.

So again, PCS is a much better predictor of 3A GOE than 3A height/distance/landing speed.

Also they should measure the airtime during the jump, that would be a rudimentary proxy for the rotation speed. Although I am not sure personally that it should be counted as important for the judges, since in the end number of completed revolutions is what counts.

I'm pretty sure jump height is a direct proxy for air time, if I remember my kinematics correctly. I'll go look up the equations later. I'm not sure air time is a good proxy for rotation speed though, because some skaters have a delay in rotation and some skaters finish their rotation earlier than others before landing.

I’d be curious to know if there is any data on rotation speed and GOE? I’m wondering if judges are actually using this data when judging. I’d imagine that’s perceptually easier and more intuitive for someone to distinguish differences very quickly rather than something like height or distance of a jump which seems to require technological assistance.

Also, it’s been a long time since I’ve taken physics but doesn’t someone have to sacrifice height for distance and vice versa? I’d think you’d need to take both data points into account. Someone jumping high but not far or far but not high may not be receiving scores as high as someone who is utilizing both height and distance.

No, height and distance actually have a very small positive correlation, so higher jumps tend to also have bigger distance (at least in this data set), though this tendency is pretty weak. It makes sense though--more height=more are time=more time to travel across ice. Although I suppose athletes do have a choice between how much force they want to use to spring upward versus across, which might be what you're thinking of and may explain why the correlation is so weak.

 

[CanadianSkaterGuy]

I don't see why the size of the skater should affect what is objectively considered a big jump--in sports like long jump, they don't score you differently based on how tall you are. In fact, I think that that actually speaks to a flaw in letting judges determine this subjectively--a jump may *look* higher because of the size of the skater, but not actually be higher.

I also think that, while yes, the correlation between GOE and size of jump should not be exceptionally high because they are not the only components of a jump's quality, the fact that there is virtually no correlation at all is concerning. Again, you are much better off guessing GOE based off PCS and BV than you are based on ice scope jump statistics--like, absurdly so. The relationship between PCS and GOE, and the relationship between BV and GOE, is quite strong, whereas the relationship between jump size and GOE barely exists. Right now, jump size explains somewhere in the neighborhood of 1% of the variation in GOE, 0% if we are considering height. That seems incredibly low.

In fact, let me re-iterate that if you remove Yuzuru Hanyu from the data set because he's a large outlier, there is actually a negative relationship between jump height and GOE. Essentially, if your name is not Yuzuru Hanyu, jumping higher is correlated with lower scores.

Yes, of course this analysis depends on the accuracy of the ice scope data. Having seen multiple measurements of the same skater jumping the same jump on different, the numbers seem fairly consistent and do not vary wildly, which seems to me to provide at least some evidence of its reliability. Additionally, the numbers seem quite plausible--we see, for instance, that Shoma doesn't jump very high, but he gets good distance and has very high speed exiting the jump, which matches the data given. Or Kolyada jumping high but not far, Yuzuru's 3A being gigantic, and so on. I'm certainly more inclined to trust it than fan made estimates using body-lengths or board-comparisons, which depend a lot on the perspective the shots are taken and are prone to being colored by fan bias.

In long jump the jump is objective so the height or size of the athlete doesn't matter. But if you're talking about a subjective GOE bullet like "very good height and distance" then that is to be taken into consideration. If the GOE bullet was "height of at least 0.6 m and distance of at least 3.00 m" then it would be a comparable analogy but like I said judges aren't pulling out measuring tapes or looking at the axel cam -- they are making a subjective call.

 

[Sugar Coated]

I'm going to have to familiarize myself with more complicated regressions more, but it turns out that height, distance, and landing speed aren't all that correlated, so multicollinearity isn't a significant problem. But the multiple linear regression I ran suggested the same thing as the initial analysis--none of the 3 variables are significant predictors of GOE. In fact, the calculator I ran just refused to include height and distance in the model at all due to the lack of effect of those two quantities on GOE I think it went for landing speed just because out of the three the correlation between it and GOE is the strongest by a hair. So the model just reduces down to the single-variable model for landing speed vs GOE.

So again, PCS is a much better predictor of 3A GOE than 3A height/distance/landing speed.




I'm pretty sure jump height is a direct proxy for air time, if I remember my kinematics correctly. I'll go look up the equations later. I'm not sure air time is a good proxy for rotation speed though, because some skaters have a delay in rotation and some skaters finish their rotation earlier than others before landing.



No, height and distance actually have a very small positive correlation, so higher jumps tend to also have bigger distance (at least in this data set), though this tendency is pretty weak. It makes sense though--more height=more are time=more time to travel across ice. Although I suppose athletes do have a choice between how much force they want to use to spring upward versus across, which might be what you're thinking of and may explain why the correlation is so weak.

It’s been a while with physics and statistics. But air time is related to height and I’d assume speed (entry speed?) is related to distance. My point was not necessarily that these were inversely correlated, but rather when you account for the combination of height and distance you find a correlation where neither exists independently. Is this a mediation analysis? Again, it’s been way too long since statistics classes so correct me if I’m wrong.

 

[elysium]

I'm pretty sure jump height is a direct proxy for air time, if I remember my kinematics correctly. I'll go look up the equations later. I'm not sure air time is a good proxy for rotation speed though, because some skaters have a delay in rotation and some skaters finish their rotation earlier than others before landing.

The body rotation, among a lot of other things, is affecting the air resistance, which may make things more complicated than the standard equations. I don’t know how large the impact is, could be just secondary or maybe a lot. I agree though that the delay in rotation is yet another complication.

 

[Shanshani]

It’s been a while with physics and statistics. But air time is related to height and I’d assume speed (entry speed?) is related to distance. My point was not necessarily that these were inversely correlated, but rather when you account for the combination of height and distance you find a correlation where neither exists independently. Is this a mediation analysis? Again, it’s been way too long since statistics classes so correct me if I’m wrong.

Hm, okay, I get your point. The effect size of both height and distance is so non-existent that I'm a bit skeptical there will be anything though--there certainly wasn't enough to support a standard multi linear regression. This is a little bit out of my depth though, so I will have to do more research.

The body rotation, among a lot of other things, is affecting the air resistance, which may make things more complicated than the standard equations. I don’t know how large the impact is, could be just secondary or maybe a lot. I agree though that the delay in rotation is yet another complication.

I guess that's true. I was initially inclined to discount air resistance because the hang time isn't that long even on the biggest jumps, but now that I think about it air resistance could easily have a significant effect on the skater's rotation. Still not sure there's much of an effect in terms of up-down motion though.

 

[Shanshani]

In long jump the jump is objective so the height or size of the athlete doesn't matter. But if you're talking about a subjective GOE bullet like "very good height and distance" then that is to be taken into consideration. If the GOE bullet was "height of at least 0.6 m and distance of at least 3.00 m" then it would be a comparable analogy but like I said judges aren't pulling out measuring tapes or looking at the axel cam -- they are making a subjective call.

I don't understand this argument. This is a sport, is it not? Why should the size of your jump as seen through the eyes of potentially biased judges matter more than the actual size of your jump? We have the technological equivalent of tape measures now--we don't need to rely on a panel of unreliable human beings, who, like all human beings, suffer the cognitive biases that our species is plagued with, like reputational bias, tendency to conformity, and so on. At least not for this component of jump quality, anyway.

A lot of what leads people to not take figure skating seriously as a sport has to do with the subjective nature of judging. Of course, some of figure skating is subjective, so that will never be entirely eliminated, but quite a lot of figure skating is objective (how fast, how high, how long, how many degrees of rotation, etc.) and the more we can assess those objective components objectively, the more seriously figure skating will be taken as a sport.

 

[elysium]

I guess that's true. I was initially inclined to discount air resistance because the hang time isn't that long even on the biggest jumps, but now that I think about it air resistance could easily have a significant effect on the skater's rotation. Still not sure there's much of an effect in terms of up-down motion though.

I think the logic should be the following. Both the distance and the height are functions of both the velocity vector and air time. We want to solve for the air time given the distance and height. Factors like shape of the body and rotation will change the velocity due to the air resistance. The air time must therefore be affected by the air resistance. In other words, given the same distance and height, people managing the air resistance differently must have different air time.

Anyway, interesting discussion. Running regression against country might have higher correlation .

 

[sanfan]

Also, GOE reflects things that can be measured objectively (i.e. height, distance, speed into and out of the jump) and those that cannot (landing quality, difficulty of entry, air position, etc.). I don't know how you'd standardize for the latter.

I don't think it's a problem at all. We can easily separate GOE into 2 parts: one part is scored by computer and another part is scored by the judges. It's a simple math and not difficult to calculate at all, just like how they add components together.

Right now the rule is designed to have these measurable elements (height, distance) and other elements (landing quality, difficulty of entry) together simply because they are evaluated by the same judges, not because they cannot be measured separately. I feel our human brains tend to be limited by some existing forms.

 

[drivingmissdaisy]

I don't think it's a problem at all. We can easily separate GOE into 2 parts: one part is scored by computer and another part is scored by the judges. It's a simple math and not difficult to calculate at all, just like how they add components together.

Right now the rule is designed to have these measurable elements (height, distance) and other elements (landing quality, difficulty of entry) together simply because they are evaluated by the same judges, not because they cannot be measured separately. I feel our human brains tend to be limited by some existing forms.

Sorry, what I was getting at is that we cannot currently look at GOE scores and reliably determine what impact height, distance, or speed have on the mark given. You could have a jump with great height and distance that ends with a fall, and that -5 score would not reflect those positive qualities. We also see a lot of jumps that earn high GOE solely on the strength of subjective measures like varied air position and difficult entry, and this would look like an outlier if it earned a high score without being exceptional on any objective measures.

 

[CanadianSkaterGuy]

I don't understand this argument. This is a sport, is it not? Why should the size of your jump as seen through the eyes of potentially biased judges matter more than the actual size of your jump? We have the technological equivalent of tape measures now--we don't need to rely on a panel of unreliable human beings, who, like all human beings, suffer the cognitive biases that our species is plagued with, like reputational bias, tendency to conformity, and so on. At least not for this component of jump quality, anyway.

A lot of what leads people to not take figure skating seriously as a sport has to do with the subjective nature of judging. Of course, some of figure skating is subjective, so that will never be entirely eliminated, but quite a lot of figure skating is objective (how fast, how high, how long, how many degrees of rotation, etc.) and the more we can assess those objective components objectively, the more seriously figure skating will be taken as a sport.

Yes, a lot of figure skating is objective, but among all the other data that has to be compiled that goes into a judge's GOE, the amplitude of a jump is (currently) less significant than something like sufficient rotation, or calling the level of a spin/footwork. All these things have to be measured and if jump amplitude is yet another thing thrown into the mix/wait time for scores.

Unless the specific height becomes a variable that goes into the final score of the jump -- in trampoline gymnastics, airtime is actually considered as part of the score, but that to me is an integral part of the sport. Amplitude on jumps is just one aspect of a tons of things that go into a performance, and so taking the time to score more objectively based on jump height/distance/landing speed is more accurate, sure... but it's drawing attention to such a trivial thing. Like what's supposed to happen - are there going to be minimums set on jump height/distance and a green light goes off if a jump manages to hit it? And then that contributes to ONE GOE bullet being given or not given? Like, that's kind of ridiculous/expensive effort for a relatively trivial consequence on scoring. Now, if a jump's GOE was 50% based on amplitude then more of a case could be made for accurately assessing the amplitude, but again it's unfair -- a skater who is 5 feet tall won't necessarily jump as high as a skater who is 6 feet tall (or who knows, maybe they will since they're lighter and it's easier to throw lesser body weight higher into the air)... all I'm saying it's not so simple to say someone's a better jumper simply because their jump goes higher off the ice.

If a skater is 3 feet tall and jumps 3 feet in the air, that's more impressive than a skater who is 8 feet tall and jumps 4 feet in the air.

Maybe there should be a jump bonus at the end - like, based on the ratio to your height, if you manage to execute your 7 jumping passes collectively X metres in total height, X metres in total distance, and X meters in landing speed, and you get an extra point(s) if you reach that threshold.

Let's be honest, Japan during Worlds is a tad more hardcore, and not every place is going to be able to afford/implement the Axel cam every time.

 

[CanadianSkaterGuy]

I'm also skeptical of the accuracy of the ice scope:

Look at Eunsoo Lim's 2A - https://www.youtube.com/watch?v=uqivP8vP2nM&t=3m20s

The landing speed is 19.5 m/s. I remember watching it and being like, wait a minute, that's faster exit speed than any of the men's triple axels.

Now look at Shoma Uno's 3A (landing speed of 18.3 m/s).
https://www.youtube.com/watch?v=w3gJbfxTSis#t=2m20s

I'm not sure when/how precisely the landing speed is measured, but it looks like Uno's got way more speed on the landing of his 3A than Lim has on her 2A.

 

[cohen-esque]

Really the distance on the jump cam seems the most suspect to me. A lot of the skaters could probably get a meter or two taller just by laying down on the ice.

 

[Sk8swan]

The human eye is excellent at perceiving a jump's height, distance and speed, and combining it instantly with other perceptions of the jump like rotation, air position and speed ... plus the other characteristics you mentioned like how well a jump expresses the music, tempo, tone, mood. Replacing the human factor ever with machines wouldn't be the way to go.

I really enjoyed the graphics, though. It was interesting, like diagramming a sentence is interesting. Not to measure up whether the judging was done right or not, but as a demonstration of the mechanics. Also, it gives perspective on just how high, far, and fast a jump travels ... something that can be somewhat distorted by watching video. I sometimes stop a replay in the middle of a pairs throw jump, just to help my eye measure what I'm looking at. Seeing the differences, such as the effect of a tall pairs girl like Ashley Cain being thrown to where her feet are at the level of her partner's waist, is kinda fun. So now I pay more attention to the overall effect of a jump, and I notice when the girl's feet only reach to her partner's groin level. I would say I enjoyed the graphics and information on the pairs throw jumps more than the singles.

Me too! I thought the pairs throws graphics were very interesting indeed! Watching on video and not live can be totally misleading of how big in terms of length and height a throw can be! Sui&Han throw 3sal in the LP looks huge if you watch it during the program but when you see the graphics you find out it's not as big. It was actually smaller than let's say DM&G who are never addressed as pair with the best throws but they do have the throws that, in terms of lenght, travel the most (lenght second only to P&J in the LP). I think the fact Han is so small makes your eye believe that their throw is immense. I cannot really make my mind on whether these stats should be used to judge elements expecially throws. It's obvious for Han is an exceptional effort to launch his partner the way he does and it should be 'easier' for a guy like Guarise; on the other hand for the girl to be thrown in a smaller jump is definitely easier to control the landing. Interesting topic though!