Youth Soccer Rankings ?

Congrats on a successful season! Sounds like the top few clubs in bracket were pretty similar in strength. There is no magic in the app, it's just a math problem, to determine which team is most likely to win the very next match based on all of their performances up to that instance. You don't get a bonus for a "win", a penalty for a "loss", or anything else concrete based on any single game outcome. Teams start to get rated better as they beat (or even play similarly to) better teams. Teams start to get rated worse if they don't do as well as similar teams.

Think about it this way. They have a huge pile of data, game results for all teams for the past year. Take all game results from 52 weeks ago to 2 weeks ago. Take all the game results from the past 2 weeks. Transform all of the old data with various weightings and rules - into predicting the actual results that happened over the past 2 weeks. Keep messing with the weightings and rules, until you optimize the number of predicted wins matching actual wins that you get over that 2 week stretch.

Now take all those weightings and rules - and start predicting the future. That's all the ratings are. A higher-rated team will beat a lower-rated team quite often, roughly 5 out of 6. Doesn't mean it will get every game correct, doesn't mean it will predict the outcome in all cases, doesn't mean a team can't have wildly different results for a period of time. But over time - its rating is going to match the performance that a team displays - it's what the data is directly based on. Its predictions are going to be better than any human looking at a single standings bracket.
 
The way I think about the ranking app is that it predicts a likely outcome and is most accurate right after the games are played and the data is updated. *

The more time that occurs between the last time two teams games are played the predicted result tends to be less valuable. This is because things the ranking app doesn't track / account for (coach leaves the team, new player, injuries, sick, weather, homecoming, etc, etc) are likely to change and play into the result. But, when you play the next team the result of the changes will get folded back into expected results. I'm just referring to the amount of time between between games.

If you were able to video practices between games and use AI to track players you could combine the results with what the ranking app already provides. I'd be willing to bet there would be 90%+ accuracy at predicting games. I wonder if MLSN is already doing this.

* The fact that the ranking app weights games played against similarly ranked teams nags at me in the back of my head.
 
If the model had more information about the games to go on, in addition to just the scores, it sure seems like it would be able to make better quality predictions - sure. However accurate it is now - it seems it could be expected to be better - up until it hits some ceiling of randomness, where even if it had every single possible parameter of every game, it would still not be perfect in predicting the future; nothing can. If it's at 83% now, what is the cap - 85%? 90%? 95%? It's just speculation at this point, but I have a hunch that it's not as far from the highest it could get as we'd think.

When people wish that it gave more credit for this, less credit for that, changed this weighting this way, or that way, because they think it would be more predictive - they don't realize that *the app is already doing that*. All of those permutations, any of them that we can reasonably imagine, are all applied to many, many thousands of game results - to see how well it can predict actual game results (defined as "predictivity", how many times does a higher rated team beat a lower rated team). Looking at the results of one team and thinking "it should weight this more", isn't particularly helpful, or accurate. Same for 5, or 10, or 20 teams. Look instead at the results of 10,000+ teams, and see if that weighting causes more successful predictions, or less successful predictions - and use the parameters that get the highest number of successfully predicted wins. It won't be consistent forever - so periodically keep checking predictions vs actual, and morph the weightings over time to keep predictivity as high as possible.
 
How does soccer ranking handle teams that play up a year? I noticed a few teams that regularly play up and get big ranking boost by beating flight 3 teams in the older age group. Then when they actually play their own age group they sucked.
 
Jamisfoes said:
How does soccer ranking handle teams that play up a year? I noticed a few teams that regularly play up and get big ranking boost by beating flight 3 teams in the older age group. Then when they actually play their own age group they sucked.
Click to expand...
It doesn't really matter which teams play each other. The ranking app defines an expected number of goals for each team from the combined histories of the teams they've played previously. If the teams exceed the expected number of goals they'll go up in ranking. If they get less than the expected number of goals they go down in ranking.
 
There's no difference whether the games are different ages, or even across different genders. It doesn't matter. It's always this collection of game results vs. the opposing collection of game results. Then the ratings are sorted by age/geography/gender in the various views - but that's secondary to the ratings themselves being created. Predictivity for cross-age matches is actually slightly better than for games vs teams of the same age. This is from an email quite awhile ago, so the exact numbers have almost certainly shifted a few percent, but at that point it was:

Screenshot 2025-08-21 091457.png

Predictivity here is defined by taking the number of games where a higher team actually beats a lower rated team, and counting that as a win. If instead a lower rated team beats a higher rated team, it counts as a loss. Ties don't count as a win or a loss. Choosing a winner for 50% of the games = .00 predictivity, it's exactly the same as a coin flip, and has zero predictive power. Choosing a winner for 100% of the games would be = 1.0 (100%) predictivity. The simple formula to go back and forth between them is you divide predictive power / 2, and add .5. Reverse it to go back the other way. It's easier to discuss "It will pick 83.4% of games correctly", than "the predictive power is .667", even if they mean the exact same thing.

So you may certainly have seen examples where a team played in a separate age, and then had wildly different results when they played their own age. But that's not indicative of the larger set of data. Playing cross ages is actually a smidgen more accurate to predict, surprisingly.
 
RandomSoccerFan said:
There's no difference whether the games are different ages, or even across different genders. It doesn't matter. It's always this collection of game results vs. the opposing collection of game results. Then the ratings are sorted by age/geography/gender in the various views - but that's secondary to the ratings themselves being created. Predictivity for cross-age matches is actually slightly better than for games vs teams of the same age. This is from an email quite awhile ago, so the exact numbers have almost certainly shifted a few percent, but at that point it was:

View attachment 30380

Predictivity here is defined by taking the number of games where a higher team actually beats a lower rated team, and counting that as a win. If instead a lower rated team beats a higher rated team, it counts as a loss. Ties don't count as a win or a loss. Choosing a winner for 50% of the games = .00 predictivity, it's exactly the same as a coin flip, and has zero predictive power. Choosing a winner for 100% of the games would be = 1.0 (100%) predictivity. The simple formula to go back and forth between them is you divide predictive power / 2, and add .5. Reverse it to go back the other way. It's easier to discuss "It will pick 83.4% of games correctly", than "the predictive power is .667", even if they mean the exact same thing.

So you may certainly have seen examples where a team played in a separate age, and then had wildly different results when they played their own age. But that's not indicative of the larger set of data. Playing cross ages is actually a smidgen more accurate to predict, surprisingly.
Click to expand...
I’m seeing a team ranked in the 30-40th a week ago, jumped to 7th within a week after a tournament against older age group in the 3rd flight.
 
Look in the full game history detail for that team. From what you shared, it's likely you'll see a bunch of green results for that tournament, meaning they overperformed by 2-3 goal differential (or more) compared to what their existing rating would predict. Their own rating would be expected to go up a smidge, and all of the teams they played (who in turn underperformed expectations) would be expected to go down a smidge. Every single game result provides additional info for the rating to be adjusted - not only that own team's direct results, but the results of all of the teams they played - and all of the teams they played, and so on.

If they keep overperforming - their rating is going to continue to increase at a decent rate. If they start underperforming - it can be expected that their rating will start taking a hit. If instead - like most teams - they continue to play at the level of their current rating, their rating will slowly and steadily increase as they age.
 
I understand what you are saying but blowing out flight 3 teams in the older age group should not give you that kind of boost in your own age group ranking.
 
Jamisfoes said:
I understand what you are saying but blowing out flight 3 teams in the older age group should not give you that kind of boost in your own age group ranking.
Click to expand...
Actually, you don't understand what I'm saying - or you wouldn't have posted the above. It betrays your thought process.

There is no concept of "flight 3" teams, or any other descriptor for any team at all. Ever. The only thing that matters in these calculations is their current SR rating. It sounds like the flight 3 teams may currently have ratings that are higher than how they are currently performing. Which means that they will start to take a hit. And the winning team is performing a bit better than expected, and will get its rating improved a bit.

But you (or I) have absolutely nothing to go on in stating "the boost in rating should be lower" or "the boost in rating should be higher". All of the rating adjustments are done at the macro level across many thousands of games to keep the predictions (and therefore the ratings), as accurate as they can be. It's like trying to use simple arithmetic to attempt to solve differential equations.
 
Soccer ranking is flawed. This goes back to the original question that started this thread. Soccer ranking gives more weigh when a team beats a similar team. A team that plays up a year is not similar to a flight 3 team playing in their own age group. But soccer ranking treats them as similar teams because their ranking after last season have them as flight 3 teams based on record. After a tryout, the A team that plays up at year got better while the flight 3 team didn’t. Now the two teams play and the younger A team blows the flight 3 team out. Soccer ranking gives the A team a big boost because it had beaten a “similar” team. And this boost gets carried over to the A team’s own age group. In reality, every A team in the younger age group got better after a tryout. The A team that played up aren’t 20 places better than before.
 
You've clearly outlined (once again), why your opinions are worthless. Something isn't true because you (or anyone else) believe it's true. It's not false because you (or anyone else) believe it's false. Figuring out what is true and how to determine for yourself whether something may be true, is a pretty good allegory for education.
 
Jamisfoes said:
Soccer ranking is flawed. This goes back to the original question that started this thread. Soccer ranking gives more weigh when a team beats a similar team. A team that plays up a year is not similar to a flight 3 team playing in their own age group. But soccer ranking treats them as similar teams because their ranking after last season have them as flight 3 teams based on record. After a tryout, the A team that plays up at year got better while the flight 3 team didn’t. Now the two teams play and the younger A team blows the flight 3 team out. Soccer ranking gives the A team a big boost because it had beaten a “similar” team. And this boost gets carried over to the A team’s own age group. In reality, every A team in the younger age group got better after a tryout. The A team that played up aren’t 20 places better than before.
Click to expand...
Don't worry about the feedback others are providing. There's many reasons why what you're describing might occur. But, the underlying math is the same for everyone. Adding weight to similar ranked games technically is evenly distributed to everyone. However in practice highly ranked teams that only play each other do get a slight boost. But this quickly goes away when they play teams outside of he small group of highly ranked teams.

Reguarding what you noticed. Everyone's perception of a huge leap in rankings is different. A highly ranked team might go up three spots and that's considered a huge jump to their parents. A middle ranked team say in the 300s might go up thirty spots overnight if they scored more goals than expected against a highly (ranked over 100) ranked team.

Also you have to keep in mind that the higher the ranking the more likely games will be variable. Lower ranked teams will be more consistant. So who knows maybe the team your team beat when playing up was on a hot streak before you beat them. Combine this with mid level ranking (300s) moves and what you've described is possible.
 
Carlsbad7 said:
Don't worry about the feedback others are providing. There's many reasons why what you're describing might occur. But, the underlying math is the same for everyone. Adding weight to similar ranked games technically is evenly distributed to everyone. However in practice highly ranked teams that only play each other do get a slight boost. But this quickly goes away when they play teams outside of he small group of highly ranked teams.

Reguarding what you noticed. Everyone's perception of a huge leap in rankings is different. A highly ranked team might go up three spots and that's considered a huge jump to their parents. A middle ranked team say in the 300s might go up thirty spots overnight if they scored more goals than expected against a highly (ranked over 100) ranked team.

Also you have to keep in mind that the higher the ranking the more likely games will be variable. Lower ranked teams will be more consistant. So who knows maybe the team your team beat when playing up was on a hot streak before you beat them. Combine this with mid level ranking (300s) moves and what you've described is possible.
Click to expand...
A model is only as good as its assumptions. Soccer ranking makes a lot of assumptions. I understand it’s a small company and they don’t have the resources to make improvements. I keep noticing it boosting teams on sparse results. It’s true the model corrects itself over time…but we don’t want those anomalies to occur in the first place.

In the end, it’s still a good program and gives you a feel where each team stands.
 
Jamisfoes said:
A model is only as good as its assumptions. Soccer ranking makes a lot of assumptions. I understand it’s a small company and they don’t have the resources to make improvements. I keep noticing it boosting teams on sparse results. It’s true the model corrects itself over time…but we don’t want those anomalies to occur in the first place.

In the end, it’s still a good program and gives you a feel where each team stands.
Click to expand...
Theres no improvements that the ranking app can make without getting into more of an analytics mindset. Which I doubt they want to do because everything will depend on personal perspectives.

But literally the math is the same for every ranking. If teams dont play each other very often and the teams they play against also dont play each other very often then the Ranking Apps prediction will be more of an educated guess. The more teams play each other and the teams they play play each other the more accurate the ranking apps predictions will be.
 
No. That hasn't been true since the first time you falsely claimed it, and it's not true now. It's not going to be true in the future.

You can repeat it until the end of time - it only reinforces the point that you're misunderstanding what you're looking at.
 
Jamisfoes said:
A model is only as good as its assumptions. Soccer ranking makes a lot of assumptions. I understand it’s a small company and they don’t have the resources to make improvements. I keep noticing it boosting teams on sparse results. It’s true the model corrects itself over time…but we don’t want those anomalies to occur in the first place.
Click to expand...

You're misunderstanding what you're seeing. The improvements you're suggesting don't fix "anomalies", they would make the ratings less likely to predict winners. You keep "noticing" things, without seeing how it fits into the bigger picture - or understanding how the changes you are suggesting would impact things that you're not considering.

Take the one you're harping on now. Team gets a boost that you think is too large for a recent game. You think - if I just could switch the boost for this team, it would make it more accurate and I would like the answer more. You can't tweak the weighting to better fit the results for one team. Or 10 teams. Or 100 teams.

What you're suggesting is that the boost should be lower for *all* games. Turns out that if they do so, they've evidently found it will make the predictions worse in aggregate. The compared ratings would predict less winners, not more.

That doesn't mean that the rules/weighting they are using now are perfect or static - they are periodically changing over time, in order to make sure that the predictivity continues to stay as high as it can be, which means that the ratings are as accurate as they can make them. They have to do that, as long as the incoming game data changes and shifts over time.
 
You must log in or register to reply here.
Back
Top