Laxnumbers Boy's National Ratings

LaxGuyWMD
Posts: 6
Joined: Tue Apr 23, 2019 8:52 am

Re: Laxnumbers Boy's National Ratings

Post by LaxGuyWMD » Thu May 02, 2019 7:18 am

Laxgramps wrote:
Wed May 01, 2019 9:45 pm
Yes but when one team beats the other head to head, has a better record and has the same strength of schedule it makes no sense that that team is ranked below the team it beat. All I’m saying is that there should be something added to the ranking algorithm to take this into account. Chaminade beat St. Anthony’s 13-7 and is undefeated. They should NEVER be ranked below St. Anthony’s.
The thing is, maybe within a conference or small (in terms of number of programs) state you could make that work, but I don't know how you'd incorporate head-to-head in what's effectively a national ranking (conference and state rankings here are simply filtered from national rankings). There's so little head-to-head that statistically it's almost meaningless. Those comparatively few situations could, though, introduce a huge bias into all the data, making differences appear larger than they really are.

To your point, looking at the data this morning (last roll-up was 3:54 AM this morning) SoS is not the same... St. Anthony's is almost 1.5 points above Chaminade. Chaminade's AGD is one higher than St. Anthony's, so the net difference is St. Anthony's being 0.5 points above Chaminade, at least as of this morning. But because of the goal differential cap, Chaminade doesn't get as much credit for the larger wins they have. That's why they're behind at this point. If you take into account those larger wins, Chaminade's AGD would be 1.5 points higher than St. Anthony's. If you remove the cap and re-roll all the data, I'll bet Chaminade comes out on top, or at the very least it's much closer than 0.5 point.

Look, you can probably find these kinds of things all over the rankings at this point. I doubt they were able to bring the data with them from LaxPower, so even if the algorithms are the same the data is all new and there's not that much of it - they effectively started from scratch. It will take more data and greater participation to sort this kind of ripple out.

If you study the math and statistics behind sports rating systems, you find that head-to-head, when it's included, is mostly used simply to make the public feel better about the ranking. It actually makes it a less reliable statistical measure of relative quality since the better team does not always win... home/away, time in the season, and other factors can affect who wins a particular game. And it gets to be even less reliable the larger the data set (number of teams/participants) is.

Laxgramps
Posts: 35
Joined: Fri Apr 05, 2019 11:38 am

Re: Laxnumbers Boy's National Ratings

Post by Laxgramps » Thu May 02, 2019 1:20 pm

LaxGuyWMD wrote:
Thu May 02, 2019 7:18 am
Laxgramps wrote:
Wed May 01, 2019 9:45 pm
Yes but when one team beats the other head to head, has a better record and has the same strength of schedule it makes no sense that that team is ranked below the team it beat. All I’m saying is that there should be something added to the ranking algorithm to take this into account. Chaminade beat St. Anthony’s 13-7 and is undefeated. They should NEVER be ranked below St. Anthony’s.
The thing is, maybe within a conference or small (in terms of number of programs) state you could make that work, but I don't know how you'd incorporate head-to-head in what's effectively a national ranking (conference and state rankings here are simply filtered from national rankings). There's so little head-to-head that statistically it's almost meaningless. Those comparatively few situations could, though, introduce a huge bias into all the data, making differences appear larger than they really are.

To your point, looking at the data this morning (last roll-up was 3:54 AM this morning) SoS is not the same... St. Anthony's is almost 1.5 points above Chaminade. Chaminade's AGD is one higher than St. Anthony's, so the net difference is St. Anthony's being 0.5 points above Chaminade, at least as of this morning. But because of the goal differential cap, Chaminade doesn't get as much credit for the larger wins they have. That's why they're behind at this point. If you take into account those larger wins, Chaminade's AGD would be 1.5 points higher than St. Anthony's. If you remove the cap and re-roll all the data, I'll bet Chaminade comes out on top, or at the very least it's much closer than 0.5 point.

Look, you can probably find these kinds of things all over the rankings at this point. I doubt they were able to bring the data with them from LaxPower, so even if the algorithms are the same the data is all new and there's not that much of it - they effectively started from scratch. It will take more data and greater participation to sort this kind of ripple out.

If you study the math and statistics behind sports rating systems, you find that head-to-head, when it's included, is mostly used simply to make the public feel better about the ranking. It actually makes it a less reliable statistical measure of relative quality since the better team does not always win... home/away, time in the season, and other factors can affect who wins a particular game. And it gets to be even less reliable the larger the data set (number of teams/participants) is.
I Agree with everything you said. My point is that nobody in their right mind could think that Chaminade should be ranked behind St. Anthony’s given they are undefeated AND beat them 13-7. The algorithm used doesn’t fact a loss into the equation like it should. Baldwinsville has three losses to teams ranked well below them yet is ranked 9th in the State. St. Joes has one loss to #1 ranked (US) Culver and has played a more difficult schedule than Baldwinsville yet is ranked #12. This is the same with Victor (8-4) which lost to Brighton (8-2) yet is ranked higher. This makes zero sense.

For these rankings to be taken seriously something needs to be done to reflect these issues. Losses should have more effect on the rankings. A win is a win, whether it’s a one goal or 10 goal differential. Yes the 10 goal difference should have some weight but to have a situation where losses aren’t factored in more heavily and head to head isn’t factored in at all makes no sense.

Just trying to help here.

LaxGuyWMD
Posts: 6
Joined: Tue Apr 23, 2019 8:52 am

Re: Laxnumbers Boy's National Ratings

Post by LaxGuyWMD » Fri May 03, 2019 7:08 am

Laxgramps wrote:
Thu May 02, 2019 1:20 pm
My point is that nobody in their right mind could think that Chaminade should be ranked behind St. Anthony’s given they are undefeated AND beat them 13-7.
I would like to think I'm in my right mind. And being in my right mind, I wouldn't necessarily take one single head-to-head loss between the cumulative results of both teams (12-1 vs 10-0) and conclude that the winner is the "better team." They certainly were during that game, I don't dispute that.

I don't know the teams in question here, but is it possible that if these teams were able to play 10 or 100 times, on either neutral site or equally distributed home/away, St. Anthony's would win more of those games than Chaminade? Of course, it's not actually possible to play that many head-to-head games to establish who's "better" over an entire season, so they use statistics to estimate it.

And I'm not going to necessarily dispute that their algorithm needs some work. There is a subjectivity to statistics in selecting and weighting the metrics used to generate an estimate. But I do believe that giving additional weight to a head-to-head win, for this particular system here where there is so little head-to-head across the total number of teams in the population, is not particularly useful. You already have the goal differential for that game which indirectly gives you head-to-head. The larger the goal differential, the more weight it carries in the roll-up of all the data. But what about when there's a one goal difference?

I think we're simply going to have to agree to disagree on this. But let me provide the following example. I'm not sure many people think that the 2007-2008 Giants were "better" than the Patriots that season. The Giants only won one out of two games they played against the Pats that season... it just so happens they won the one that really mattered from a historical standpoint. But if they played 10 times, I'll bet the Pats would have won more of them, and I'll bet most people outside of NY who could look at the teams objectively would agree with that.

Laxgramps
Posts: 35
Joined: Fri Apr 05, 2019 11:38 am

Re: Laxnumbers Boy's National Ratings

Post by Laxgramps » Sun May 05, 2019 10:25 pm

Chaminade with its win over Manhasset finally moves to #1. They beat #2 and #4 by combined 14 goals yet are barely #1. They should be #1 by a lot more than they are.

LaxGuyWMD
Posts: 6
Joined: Tue Apr 23, 2019 8:52 am

Re: Laxnumbers Boy's National Ratings

Post by LaxGuyWMD » Thu May 16, 2019 8:38 am

Laxgramps wrote:
Sun May 05, 2019 10:25 pm
Chaminade with its win over Manhasset finally moves to #1. They beat #2 and #4 by combined 14 goals yet are barely #1. They should be #1 by a lot more than they are.
So last night is an excellent example of why a single head-to-head matchup is not a good metric for stuff like this.

If you want them to be "standings" to determine a league champion or something, sure, head-to-head can be important. But that's not really what the purpose of this is. It's to try and determine the best team in the nation where there are not nearly enough games to determine that with standings. Sometimes a head-to-head matchup is a decent metric, sometimes it's not. The higher the goal differential is, the better a metric it may be.

In the two games between them, Chaminade leads St. Anthony's in goal differential, so they do have a little kick. But Chaminade also now has that extra loss. But that extra loss, to an out-of-conference and out-of-state team is a good piece of data as well. If you look at common opponents, St. Anthony's beat Delbarton 13-4, but Delbarton beat Chaminade 12-9. So what does that really tell you? I'm not sure it tells you much of anything.

And that's the point of these ranking systems. In trying to determine the best team over the course of a season, you can't put too much weight in the simple win/loss of an individual game. You get that weight by considering how much that game was won or lost by, along with how much other games were won and lost by and how much your opponents won and lost by. If/until someone finds some better analytic, goal differential IS the most important metric of a single game. Which is why I think the cap should be removed. Or at the very least, a team should not be penalized if they win by more than 10 goals.

Laxgramps
Posts: 35
Joined: Fri Apr 05, 2019 11:38 am

Re: Laxnumbers Boy's National Ratings

Post by Laxgramps » Thu May 16, 2019 3:52 pm

I agree with most of what you said. However, IMHO goal differential, while relevant, is far too much weighted in these rankings. In upstate NY once a team gets up by 12 goals we play a running clock. My son’s team has had this occur in 6 of their 18 games this year. In all of those games our coach put subs in and the opposing team scored goals which lessened the goal differential. Instead of beating a team 20-3 we would win 16-8. Putting as much emphasis on goal differential in the rankings gives coaches incentive to run up the score. To me, the difference between beating a team 15-8 v 16-4 is not that great. Clearly one was better than the other on that day. To me, losses should carry more weight. A win is a win in my book whether it’s by 2 or 10. Maybe reducing the goal differential to something less that 10 and eliminating the penalization of a team that doesn’t beat its opponents by the “prescribed” goal differential would make the ranking more accurate.

Don’t get me wrong. I love what Lax numbers is doing. Just trying to give them ideas to make these rankings more accurate.

LaxGuyWMD
Posts: 6
Joined: Tue Apr 23, 2019 8:52 am

Re: Laxnumbers Boy's National Ratings

Post by LaxGuyWMD » Fri Jun 28, 2019 12:06 pm

Laxgramps wrote:
Thu May 16, 2019 3:52 pm
I agree with most of what you said. However, IMHO goal differential, while relevant, is far too much weighted in these rankings. In upstate NY once a team gets up by 12 goals we play a running clock. My son’s team has had this occur in 6 of their 18 games this year. In all of those games our coach put subs in and the opposing team scored goals which lessened the goal differential. Instead of beating a team 20-3 we would win 16-8. Putting as much emphasis on goal differential in the rankings gives coaches incentive to run up the score. To me, the difference between beating a team 15-8 v 16-4 is not that great. Clearly one was better than the other on that day. To me, losses should carry more weight. A win is a win in my book whether it’s by 2 or 10. Maybe reducing the goal differential to something less that 10 and eliminating the penalization of a team that doesn’t beat its opponents by the “prescribed” goal differential would make the ranking more accurate.

Don’t get me wrong. I love what Lax numbers is doing. Just trying to give them ideas to make these rankings more accurate.
Maybe what you do, and this might make it harder to do, but perhaps deal a little better with blowout games when the 2nd and 3rd string are put in, is when the goal differential of a game is at some threshold you decide, you simply double the first half score and use that instead of the final score. The assumption there is that starters will play until the mercy rule kicks in. That might not always be true, but I'll bet it will be true often enough to deal with most of the score issues you describe. Just a thought...

Post Reply