Youth Soccer Rankings: Disregarding >5 Point Blowouts?

Always love a good YSR debate. Part of the problem with older ages is that when the new ECNL/ECRL/GA teams are created, they lose the ranking of the team that built it's way up. Errors should be reported on the site to combine or dislocate new and old team data. Often this does not happen or the wrong team are combined. YSR does have it's own system to catch mistakes but it's rare.

As for blowouts, it is what it is. IMO, that's the biggest issue when two teams are split by 3 or 4 goals on the index. Many more times the spread is larger in the actual games.
 
Always love a good YSR debate. Part of the problem with older ages is that when the new ECNL/ECRL/GA teams are created, they lose the ranking of the team that built it's way up. Errors should be reported on the site to combine or dislocate new and old team data. Often this does not happen or the wrong team are combined. YSR does have it's own system to catch mistakes but it's rare.

As for blowouts, it is what it is. IMO, that's the biggest issue when two teams are split by 3 or 4 goals on the index. Many more times the spread is larger in the actual games.
I think your disclaimer will need to be changed soon Mars. I have a new one for you if you would like it :)
 
I agree there are deficiencies in YSR's model, but this isn't necessarily incorrect. Not all clubs treat their ECNL/ECRL teams as true A/B teams. Often they're made up of existing teams that come in as a unit, so it's not crazy that an RL team might be better than the NL. (And this is especially true for boys as RL is so new.)

May be true for boys...... would be VERY rare for girls...... a club's best teams are all ECNL
 
Always love a good YSR debate. Part of the problem with older ages is that when the new ECNL/ECRL/GA teams are created, they lose the ranking of the team that built it's way up. Errors should be reported on the site to combine or dislocate new and old team data. Often this does not happen or the wrong team are combined. YSR does have it's own system to catch mistakes but it's rare.

As for blowouts, it is what it is. IMO, that's the biggest issue when two teams are split by 3 or 4 goals on the index. Many more times the spread is larger in the actual games.
I would actually suggest YSR limit the data set to 12 months. The purpose of the rankings is to see current standings, not historical. The older data also may poison the whole ranking if that team does a lot of cross-league play.

YSR's inability to predict blowouts is kind of my point. The inaccurate 3-4 point predicted differential appears to be a symptom of capping blowouts over 5 goals - even if it is persistent.

And please don't reply Mars if you still have that headache of a signature. I like the sentiment but I may have to block it to be able to read anything you post in.
 
I would actually suggest YSR limit the data set to 12 months. The purpose of the rankings is to see current standings, not historical. The older data also may poison the whole ranking if that team does a lot of cross-league play.

YSR's inability to predict blowouts is kind of my point. The inaccurate 3-4 point predicted differential appears to be a symptom of capping blowouts over 5 goals - even if it is persistent.

And please don't reply Mars if you still have that headache of a signature. I like the sentiment but I may have to block it to be able to read anything you post in.
How do you propose to include blowouts?

Suppose the Pink Flamingos rack up another 12-0 win over a team with a score of 19. Do you want to put a 31 into the average?

I’d say the Flamingos are in for a rough day if they ever have to play a team whose 31 came from close games.
 
How do you propose to include blowouts?

Suppose the Pink Flamingos rack up another 12-0 win over a team with a score of 19. Do you want to put a 31 into the average?

I’d say the Flamingos are in for a rough day if they ever have to play a team whose 31 came from close games.

Yes, average it in, and deal with outliers differently. I mentioned previously dropping any game which is way outside of the predicted range (say, 2 standard deviations). The problem comes when you presume blowouts are outliers when, between certain teams, they are the norm.

Also, I think you would probably see a larger spread in the point system for YSR. The top team would be more like 40 rather than 38. The bottom team in an age group would be more like 20. Right now, because of the 5 point cap, I suspect, The spread is more like 10 points rather than 20.

Sparkling Pink Flamingos was literally one of my daughter's teams. How dare you.
 
Yes, average it in, and deal with outliers differently. I mentioned previously dropping any game which is way outside of the predicted range (say, 2 standard deviations). The problem comes when you presume blowouts are outliers when, between certain teams, they are the norm.

Also, I think you would probably see a larger spread in the point system for YSR. The top team would be more like 40 rather than 38. The bottom team in an age group would be more like 20. Right now, because of the 5 point cap, I suspect, The spread is more like 10 points rather than 20.

Sparkling Pink Flamingos was literally one of my daughter's teams. How dare you.

Averages don’t help when you have a team as dominant as the Flamingos.

How do you detect an outlier when the norm is blowouts? The Flamingoes won by an average of 8 points last season. The outlier was the 4-3 loss against Purple Unicorns. If we throw out outliers, we should exclude the Unicorns game.
 
Averages don’t help when you have a team as dominant as the Flamingos.

How do you detect an outlier when the norm is blowouts? The Flamingoes won by an average of 8 points last season. The outlier was the 4-3 loss against Purple Unicorns. If we throw out outliers, we should exclude the Unicorns game.

You are dangerously close to my DD's other AYSO team. Are you my wife? If so, you are right and I am wrong.

You don't detect outliers by just raw score differentials. That is what I am advocating against. An outlier is would be determined by how far off from the predicted outcome the game is - the predicted outcome based on YSR's total aggregate data and algorithm. In your scenario, the 4-3 loss could be an outlier if YSR has Purple Unicorns at 24 and Pink Flamingos at 28 (it expected Flamingos to win by 4).

If it happens once, outlier. If it happens twice, you move the bell curve over and include both data points.
 
You are dangerously close to my DD's other AYSO team. Are you my wife? If so, you are right and I am wrong.

You don't detect outliers by just raw score differentials. That is what I am advocating against. An outlier is would be determined by how far off from the predicted outcome the game is - the predicted outcome based on YSR's total aggregate data and algorithm. In your scenario, the 4-3 loss could be an outlier if YSR has Purple Unicorns at 24 and Pink Flamingos at 28 (it expected Flamingos to win by 4).
if you have 8 blowouts and one 4-3 game, the YSR algorithm will gradually keep raising the team rating. Eventually, it has them at 31- supported by nothing but the blowouts. (We threw out the 4-3 game, because it didn’t match predictions.)

That’s a problem for the Flamingos. I don’t think they’re quite ready for Supercopa or Dallas Invitational just yet. But their rating says they are.
 
if you have 8 blowouts and one 4-3 game, the YSR algorithm will gradually keep raising the team rating. Eventually, it has them at 31- supported by nothing but the blowouts. (We threw out the 4-3 game, because it didn’t match predictions.)

That’s a problem for the Flamingos. I don’t think they’re quite ready for Supercopa or Dallas Invitational just yet. But their rating says they are.

If they decide their first time playing a high level team must be in Dallas, so be it. Once they go, those teams will beat them by 9 points each. The higher level teams will have 40 points. Flamingos are 31. The other teams in Flamingo's league are 23.
 
If they decide their first time playing a high level team must be in Dallas, so be it. Once they go, those teams will beat them by 9 points each. The higher level teams will have 40 points. Flamingos are 31. The other teams in Flamingo's league are 23.
Not quite. When they went to Dallas, they were scheduled against teams ranked 30-32. It was predicted to be close, but they lost every game 8-0.

That means every game in Dallas was an outlier. They didn’t match predictions, so we threw those games out. Flamingos are still listed at 31. (oops)

See the problem? Flamingos are a fake 31. They’re a 26, but all offense. They can beat any 23 by huge margins. Against a 25-27, they have a close game. Against a real 31, they get crushed.

The algorithm need to converge anyway. Ideally, it converges to something near 26, because those are the good games.
 
I would actually suggest YSR limit the data set to 12 months. The purpose of the rankings is to see current standings, not historical. The older data also may poison the whole ranking if that team does a lot of cross-league play.

YSR's inability to predict blowouts is kind of my point. The inaccurate 3-4 point predicted differential appears to be a symptom of capping blowouts over 5 goals - even if it is persistent.

And please don't reply Mars if you still have that headache of a signature. I like the sentiment but I may have to block it to be able to read anything you post in.
Agree. Plus team rosters can change significantly a year later. Data from a team that has a nearly completely different roster is not accurate.
 
Not quite. When they went to Dallas, they were scheduled against teams ranked 30-32. It was predicted to be close, but they lost every game 8-0.

That means every game in Dallas was an outlier. They didn’t match predictions, so we threw those games out. Flamingos are still listed at 31. (oops)

See the problem? Flamingos are a fake 31. They’re a 26, but all offense. They can beat any 23 by huge margins. Against a 25-27, they have a close game. Against a real 31, they get crushed.

The algorithm need to converge anyway. Ideally, it converges to something near 26, because those are the good games.

I don't know if this is what you are getting at, but I would agree that goals are not a stable currency. Let me put it this way:

D1 is +3 to F1
F1 is +8 to F2

What is D1 to F2's differential then? Logically, you would think +11. But, in head-to-head, because goals come easier when D1 plays against F2, it is probably more like +13 or +14. So, yes, D1 could theoretically rank higher by playing F2 all year rather than F1. But I don't think this justifies a +5 cap on GD. Goals after the first 5 in blowout games are cheaper, but not worthless.

Maybe the only way to get a reliable GD ranking is for all the teams to each play the bottom team once. Whichever team scores the most goals is the best one. Presumably SD Surf ECNL is a plus 30.
 
All good points here. The one thing I'd say in favor of a cap is that once a team is up by 4 or 5 goals, some coaches will take their foot off the pedal and have the players work on specific things (usually possession) over scoring because 1) they can and 2) it doesn't do anyone any good to roll over a lessor opponent. Teams that don't do this will end up with higher YSR ratings than teams at the same level that do do this.
 
Waiting to see how long a thread goes about an algorithm judging a bunch of 9 year olds. Do we think YSR takes "stayed up past bedtime to watch a new episode of their favorite Disney+ show" into consideration? Come on folks.
 
Waiting to see how long a thread goes about an algorithm judging a bunch of 9 year olds. Do we think YSR takes "stayed up past bedtime to watch a new episode of their favorite Disney+ show" into consideration? Come on folks.
You mean I will be any less ridiculous when my kid turns 15?

Don't bet on it.

Go Flamingos!!!!!!
 
Waiting to see how long a thread goes about an algorithm judging a bunch of 9 year olds. Do we think YSR takes "stayed up past bedtime to watch a new episode of their favorite Disney+ show" into consideration? Come on folks.

Considering that the process of gathering and manipulating the data is probably 95% automated, no.
 
Back
Top