Practice Reports - Friday, October 26

[Metis]

Don't count 100% on my PBP being accurate as to which jump - I used a combination of planned program and visuals (I'm still learning how to tell between them!) but either way, the 3F is his first planned jumping pass.

I'd predict Canadian polite...:think:

Hey, I very much appreciate the PBPs, and when it comes to Lutz/flip, I can ID them, but with the way edge warnings and calls have been handed out like Halloween candy lately, every one is an immediate slo-mo to see if the tech panel might have something to feast on. [emoji23]

Jason’s practice was uploaded — looking solid, definitely an improvement from ACI (which is to be expected but good to see), and the 3A looked very slightly nervous but he landed it. It’s going to be weird not holding my breath during Nathan and Jason’s 3As....

And, yeah, Shoma has some weird inverse correlation magic going with his practice performance and competition performance, or so the legend goes. Someone should crunch numbers on how much of a leading indicator practice run-throughs are... *looks at self*... if anyone needs me I’ll be in my room

 

[tabsy]

Hey, I very much appreciate the PBPs, and when it comes to Lutz/flip, I can ID them, but with the way edge warnings and calls have been handed out like Halloween candy lately, every one is an immediate slo-mo to see if the tech panel might have something to feast on. [emoji23]

Jason’s practice was uploaded — looking solid, definitely an improvement from ACI (which is to be expected but good to see), and the 3A looked very slightly nervous but he landed it. It’s going to be weird not holding my breath during Nathan and Jason’s 3As....

And, yeah, Shoma has some weird inverse correlation magic going with his practice performance and competition performance, or so the legend goes. Someone should crunch numbers on how much of a leading indicator practice run-throughs are... *looks at self*... if anyone needs me I’ll be in my room

It's an interesting question. Maybe when I have less work on I'll attempt it (my maths skills are a little rusty but used to be decent...I hope hwell

I've got enough to do this weekend - I'm having my first go at scoring (Pairs), and I'm doing it for the next two weekends as well!

Jason is looking good. Still a little worried about the FS, he seems a lot more unstable/unsure about it.

Hopefully Shoma's new costume will bring him all and any luck. I like it a lot

 

[Arriba627]

Shoma:

SP

YES to the purple velvet!

I think avoids doing 4F. Long pause.

Comes back in when the music changes tempo. Some of steps, all quite low energy.

Unless I missed something, he didn't do a jump.

Outside of his run thru, lots of jumps. Looked good. Gorgeous costume. Go Shoma!

 

[Colonel Green]

The best banner in the arena.

 

[tabsy]

Colonel Green, that link didn't work.

Don't suppose it was JUNLIIIEEEETTTTT by any chance? Brilliant banner

ETA - works now - I was right!

 

[SnowWhite]

It's the little Conrad Orzel face that really makes it for me.

 

[tabsy]

It's the little Conrad Orzel face that really makes it for me.

I hadn't even spotted that

 

[Metis]

It's an interesting question. Maybe when I have less work on I'll attempt it (my maths skills are a little rusty but used to be decent...I hope hwell

I've got enough to do this weekend - I'm having my first go at scoring (Pairs), and I'm doing it for the next two weekends as well!

Jason is looking good. Still a little worried about the FS, he seems a lot more unstable/unsure about it.

Hopefully Shoma's new costume will bring him all and any luck. I like it a lot

We’d have to build a dataset based on PBPs from practice threads, while also cross-checking footage whenever possible to measure intercoder reliability, which is going to be a mess since PBPs aren’t all done by one user and not all practice footage still exists, so there’s going to be inherent error in the user-reported data. For cases of known injury and illness, we’d want to downweight or remove those from the sample, especially in instances in which a skater doesn’t complete practice for some reason and/or WDs. Also, we can only compare what’s done in runthrough to competition, and most skaters don’t do their programmes from start to finish when their music plays — they tend to skip a few elements, especially the StSq and spins.

More fun: some issues flagged by the TP may not be visible from practice cam footage as well, which will make “success in practice” harder to determine (especially if the skater has periods of being off-camera...). And, finally, when coding a dataset for practices, we’d have to establish exactly what we’re counting as “practice” in the comparison: what’s executed when the skater’s music is on and it’s their turn or reports covering the entire practice session? Because when people discuss the gap between practice and performance, we hear about Bradie hitting 3Lz-3Lo a half-dozen times in practice, but those six combos weren’t all performed during the time reserved for her music. (We’d also have to decide if the 6 minute warm up counts.) Determining the extent to which practice is a leading indicator both overall and for specific skaters is just a basic regression analysis, with the null hypothesis being that there’s no meaningful relationship between the two... and you could p-hack your way into any result you desire depending on how expansive your practice dataset is; how elements are coded (lots of potential false positives in the practice data, and also from protocols depending on the tech panel); how you count no-shows/illness/injury/etc.

(Yes, I took courses in research methodology. No, don’t run away!)

tl;dr meaningful statistical analysis in skating is relatively difficult due to a few variables — technical panels not issuing calls is a huge problem (most of Medvedeva’s Lutzes would be considered a full success due to the lack of even a ! call, for example); the ISU’s own rules being applied in pieces (jumps with visible UR issues being given +3 in the old system, etc.); timing (early in the season or late? No one expects much at ACI), and we have enough data on some skaters to be able to test to see if variance in practice and performance is explained by “Hanyu always Hanyus ACI” versus “this is actually a meaningful data point”; skaters integrating new jumps (Rizzo, for example, and the 4T). It’s pretty difficult to control for all the inherent randomness in skating. (I’ve maybe thought too much about this.)

 

[tabsy]

If anyone is interested, Piper appears to have swapped out the yellow dress for a very delicate blue one, with a covered but sheer back, and a very low sharp v neck.

 

[Colonel Green]

If anyone is interested, Piper appears to have swapped out the yellow dress for a very delicate blue one, with a covered but sheer back, and a very low sharp v neck.

I think that’s her tango FD dress from 2016-17.

 

[tabsy]

We’d have to build a dataset based on PBPs from practice threads, while also cross-checking footage whenever possible to measure intercoder reliability, which is going to be a mess since PBPs aren’t all done by one user and not all practice footage still exists, so there’s going to be inherent error in the user-reported data. For cases of known injury and illness, we’d want to downweight or remove those from the sample, especially in instances in which a skater doesn’t complete practice for some reason and/or WDs. Also, we can only compare what’s done in runthrough to competition, and most skaters don’t do their programmes from start to finish when their music plays — they tend to skip a few elements, especially the StSq and spins.

More fun: some issues flagged by the TP may not be visible from practice cam footage as well, which will make “success in practice” harder to determine (especially if the skater has periods of being off-camera...). And, finally, when coding a dataset for practices, we’d have to establish exactly what we’re counting as “practice” in the comparison: what’s executed when the skater’s music is on and it’s their turn or reports covering the entire practice session? Because when people discuss the gap between practice and performance, we hear about Bradie hitting 3Lz-3Lo a half-dozen times in practice, but those six combos weren’t all performed during the time reserved for her music. (We’d also have to decide if the 6 minute warm up counts.) Determining the extent to which practice is a leading indicator both overall and for specific skaters is just a basic regression analysis, with the null hypothesis being that there’s no meaningful relationship between the two... and you could p-hack your way into any result you desire depending on how expansive your practice dataset is; how elements are coded (lots of potential false positives in the practice data, and also from protocols depending on the tech panel); how you count no-shows/illness/injury/etc.

(Yes, I took courses in research methodology. No, don’t run away!)

tl;dr meaningful statistical analysis in skating is relatively difficult due to a few variables — technical panels not issuing calls is a huge problem (most of Medvedeva’s Lutzes would be considered a full success due to the lack of even a ! call, for example); the ISU’s own rules being applied in pieces (jumps with visible UR issues being given +3 in the old system, etc.); timing (early in the season or late? No one expects much at ACI), and we have enough data on some skaters to be able to test to see if variance in practice and performance is explained by “Hanyu always Hanyus ACI” versus “this is actually a meaningful data point”; skaters integrating new jumps (Rizzo, for example, and the 4T). It’s pretty difficult to control for all the inherent randomness in skating. (I’ve maybe thought too much about this.)

I'm impressed- that's all absolutely fascinating. I'm not running away, but I am all the more unaware of how little I know about methodology that seems like enough complications and variables to take into account you could do an entire Masters thesis on it :shocked: I think given the difficulty in accessing past streams of practices, and not having a 'fancam' (not sure if that's the right word) of all the skaters all the time on the ones we do have, if anyone was ever going to attempt it, it'd be easier to start a dataset from this point onwards, especially with the new rules and changes, rather than use historical data?

Thank you for your response - I enjoyed skimming through and will read it properly later tonight (maybe during some zamboni time)!

- - - Updated - - -

I think that’s her tango FD dress from 2016-17.

Ah! She's probably using it to make sure she doesn't damage the new one before the RD.

 

[SnowWhite]

I hadn't even spotted that

Conrad reposted a picture of the banner on his story, so I saw it there.

 

[Colonel Green]

Panenkova - fall on opening jump, 2A good, popped the second part of combination.

This can’t help but pale in comparison to Wakaba’s “Skyfall”.

 

[tabsy]

Daria P:

Skyfall

3Lo, arms above, fall
2A landed
3Lz2T I think. Lean on Lz.
Jumps shaky. Some pauses.


I need to go for a bit, but I'll be back for Group 2 if anyone wants PBP?

 

[NoNameFace]

View attachment 128

The best banner in the arena.

The banner steals the show, totally. And Conrad's cameo is hillarious

 

[Colonel Green]

Alaine had a solid run-through.

Daria’s other jump attempts in practise were a lot more successful on the whole, albeit with a heavy lean. The second half of her combination looks like a struggle, though.

 

[Nimyue]

Who thought this music was a good idea? Ugh....

 

[tabsy]

Mako Yamashita:

3Lz3T nice
2A slightly out of camera frame
3F I think

I have a feeling that Mako might be the surprise bronze medallist in this event, even if she's not in 3rd after the short...

 

[tabsy]

Mariah:

Axel popped I think
3Lz3T
3F nice

 

[tabsy]

Elizaveta Tuk:

Good 3A I think!
3T3T almost into the boards
Pop on the Lz

I have to say I kinda miss the shirt a la Lombardia