New FHSAA rankings out

[Dr. D]

On the FHSAA website, for those interested.

[Perspective]

So, just curious:  by how many points would #27 Hawthorne be favored over #31 Northwestern (neutral site, of course)? 

[THIS_IS_DILLARD]

7 minutes ago, Perspective said:

So, just curious:  by how many points would #27 Hawthorne be favored over #31 Northwestern (neutral site, of course)? 

Negative lol 

[Ray Icaza]

14 minutes ago, Perspective said:

So, just curious:  by how many points would #27 Hawthorne be favored over #31 Northwestern (neutral site, of course)? 

Or #22 Rockledge vs #43 Osceola?  Oh, nearly forgot we have that score already (30-7 in Kowboys favor).

[Perspective]

24 minutes ago, Perspective said:

So, just curious:  by how many points would #27 Hawthorne be favored over #31 Northwestern (neutral site, of course)? 

Interestingly, when MaxPreps uses their own formula, Northwestern is ranked 14 and Hawthorne is ranked 81. 

Seriously, how can there be that much difference between the two rankings?  Is this all because the FHSAA version doesn't include point differential in the formula?

[181pl]

let’s just hope it comes out correctly in the wash. I’ve looked at some of the teams and they are being rewarded for record despite playing schedules that are soft as cotton. While others have a bunch of losses and are fairly highly ranked. There has to be a better balance between the two. You shouldn’t be rewarded for over scheduling and you shouldn’t be rewarded for beating up on cotton balls. 

 

I know you folks are tired of hearing this, but this is why we need large districts. It eliminates  90% of these issues. Yes you may have one or two teams annually that may be left out they were probably better than some team in another district or region, but at the end of the day it’s the most fair way instead of some arbitrary formula.

[Dr. D]

Agreed.  Here's how easy it would be, using current 4M as an example.  No wait, a few teams would have to drive 1-2 hours, once or twice a year.  Or couldn't schedule 8 home games.  Or couldn't schedule to their program's "philosophy".  Or couldn't fit 495 different agendas into a logical approach to determining a true champion on the field, without the aid of computers in El Dorado Hills, California.  Never mind.

 

 

[Perspective]

Makes all the sense in the world Dr.D, at least with the larger metro areas.   How would it work for the smaller suburban areas?  I'd like to see a similar set-up for, say, 2S.  

[anonimis]

3 hours ago, Perspective said:

Interestingly, when MaxPreps uses their own formula, Northwestern is ranked 14 and Hawthorne is ranked 81. 

Seriously, how can there be that much difference between the two rankings?  Is this all because the FHSAA version doesn't include point differential in the formula?

The Florida rankings could be weighing out of state games differently.  I think I read they don't use previous years results as much as the national ratings.  If the teams opponents and opponents of opponents have little crossover play, their ratings aren't as comparable as they seem.

 

For what it's worth, I think they choose not to include margin of victory to stop coaches from running up the score, which they would be obligated to do.  Take the Chaminade vs Somerset Academy Key game.  If margin of victory is a factor and there is no cap, they would have scored well over 100.  Chaminade won 90-0 and while I didn't watch the game, I am sure they weren't actively trying to run up the score, but rather a national powerhouse was forced to play a startup football team in its second season that has never won a game.  Rating teams without factoring in the score actually works well, but it requires more sophisticated matchmaking than the current district/region/class system.  Matchmaking forces teams of similar scores to play each other with the winners score improving and the losers score falling.  After a few weeks teams converge around their rating (depending on the rating system, there is also a variance rating, which is helpful for matchmaking).  Pairing teams against each other by school population is nonsense in 2022.

 

Anyway, I updated my spreadsheet of weekly scores.  I added a column for risers and fallers in October and a column measuring the difference between a teams rating and their opponent rating.

https://docs.google.com/spreadsheets/d/19-MEgcJCmU7gkFDTGlT3qS_7WCDCcFHjzLC1qMicz90/edit?usp=sharing

If you notice any errors, let me know.

It's also possible to scrape all the results off of maxpreps and check the weekly ratings against the actual game outcomes.  It should get more accurate each week.  More work than I want to do right now though.

[Floridaatlantic1]

I have seen examples where teams get run off field and rise a few spots in the maxpreps rankings and other times where the drop like a rock. Using maxpreps without knowing their formula is a bad idea. The system was better when it was a third record and the. Opponents record and a third and opponents opponents record for a third but that is too easy. No money to be made off that 

[Dr. D]

58 minutes ago, Perspective said:

Makes all the sense in the world Dr.D, at least with the larger metro areas.   How would it work for the smaller suburban areas?  I'd like to see a similar set-up for, say, 2S.  

I've only put 4M and 4S on paper, but at first glance it looks workable in 2S.  I'm not as familiar with some of the smaller locales in central and southern Florida, so it will take some time to get something together (and I'm not getting paid to do the FHSAA's job, although I would settle for 10% of the MaxPreps $$$ ).  I promise I'll get back to you with an honest assessment of whether it would work at this level or not.  

[anonimis]

22 minutes ago, Floridaatlantic1 said:

I have seen examples where teams get run off field and rise a few spots in the maxpreps rankings and other times where the drop like a rock. Using maxpreps without knowing their formula is a bad idea. The system was better when it was a third record and the. Opponents record and a third and opponents opponents record for a third but that is too easy. No money to be made off that 

With RPI, teams had their scores go down from winning.  None of the systems are perfect, but its hard to say what the best is without the formulas, scripts, and data.  I wish there was more transparency.  Specifically, which teams are you talking about?  I'm trying to figure out which system they are using.  The one I thought they were using doesn't allow for that.

[DarterBlue2]

7 hours ago, Perspective said:

Interestingly, when MaxPreps uses their own formula, Northwestern is ranked 14 and Hawthorne is ranked 81. 

Seriously, how can there be that much difference between the two rankings?  Is this all because the FHSAA version doesn't include point differential in the formula?

Something is very fishy about such large differentials even if point margins are ignored. Several teams are going to feel like they got hosed when the final rankings shake out. Go back to RPI or go back to district champ and runner up with 6 team or greater districts. 

[PinellasFB]

10 hours ago, Perspective said:

Interestingly, when MaxPreps uses their own formula, Northwestern is ranked 14 and Hawthorne is ranked 81. 

Seriously, how can there be that much difference between the two rankings?  Is this all because the FHSAA version doesn't include point differential in the formula?

Correct.  Without margin of victory, you might as well go back to RPI.

[Perspective]

I'll ask the question another way:  is there anyone on this board that thinks Hawthorne would stay within 14 points of Northwestern in a neutral field game?   If not, how can the FHSAA defend a ranking system that has Hawthorne ranked ahead of Northwestern?  Obviously those two teams are not in the same classification, so we don't have to worry about one of those teams making the playoffs at the expense of the other.  But my point is this:   if the system is that flawed, there will be some serious issues when it comes time to select the four non-district winners that make the playoffs in each region.  

[Floridaatlantic1]

There will be issues with any system. I just have a problem with the lack of transparency of maxpreps ratings. I see a team with four losses ranked super high and they got beat bad in three of them. The latest. A 27plus beat down and they rose four spots in regular maxpreps rankings of the state. The regular rankings must focus too much on Strength of schedule for that non sense to happen 

[anonimis]

A team can go up in rank in this situation.

A team(I'll call them Alphas) loses a game to a much higher rated opponent.  The expected outcome occured.  The much higher rated team beat the much lower rated team.  This means both teams ratings don't move very much.  Now, if many teams above team Alphas loses against teams with similar or much worse ratings, their new ratings will be lower than the Alphas new rating.  This would result in a team losing, but going up in ranking.

 

I think we agree that strength of schedule is weighted highly.  I think weighing strength of schedule into a rating is very important.  I think the problem lies in not accurately measuring that strength (due to lack of matchmaking and too few meaningful games).

[Dr. D]

20 hours ago, Perspective said:

Makes all the sense in the world Dr. D, at least with the larger metro areas.   How would it work for the smaller suburban areas?  I'd like to see a similar set-up for, say, 2S.  

@Perspective, as promised, here are the current 66 teams in 2S arranged into 8 districts. This actually works better than I imagined. The longest drives are in the Panhandle, but that is already the case for teams there. All the districts are reasonably compact (except there is no good answer for Key West). This actually creates some good local rivalries, and a lot of the district games created are already being played as non-district games. Problem solved.  Don’t have to worry if it’s better to lose to a 7-3 team or beat a 3-7 team, just beat the teams in your district.  And I won’t even charge the FHSAA for my product. 

 

 

[PinellasFB]

What you hope for in any relative power index algorithm is it is generally correct and then grows more accurate as the season progresses, resolving inevitable anomalies over time.  An anomaly is one team ranked over another team that is clearly inferior to the team it is ranked ahead of.  Looking at the FHSAA-maxpreps rankings, I think this is generally true?  There are still anomalies but they are getting corrected each week.   No ranking will be perfect or even near perfect but you need one that is at least good enough.  Only time will tell if the fhsaa bastardization of the maxpreps ranking is good enough.  It's clearly inferior to the standard maxpreps rankings though. 

Perhaps instead of margin of victory, they can implement margin of defeat?   A win by 1 point or 70 counts the same, but a loss by 1 is much better than a loss by 70.  At least we can reward teams for playing a tough schedule with a bunch of close losses.  

[Perspective]

34 minutes ago, PinellasFB said:

 

Perhaps instead of margin of victory, they can implement margin of defeat?   A win by 1 point or 70 counts the same, but a loss by 1 is much better than a loss by 70.  At least we can reward teams for playing a tough schedule with a bunch of close losses.  

Dillard seconds the motion.   

[Perspective]

2 hours ago, Dr. D said:

@Perspective, as promised, here are the current 66 teams in 2S arranged into 8 districts. This actually works better than I imagined. The longest drives are in the Panhandle, but that is already the case for teams there. All the districts are reasonably compact (except there is no good answer for Key West). This actually creates some good local rivalries, and a lot of the district games created are already being played as non-district games. Problem solved.  Don’t have to worry if it’s better to lose to a 7-3 team or beat a 3-7 team, just beat the teams in your district.  And I won’t even charge the FHSAA for my product. 

 

 

Nice work Doc.  Now that you've done 25% of the classes (not counting Rural), I'm curious to see how the other six would play out, but you've done plenty of work, so I'm not asking you to do more.  Just curious. 

[Nulli Secundus]

40 minutes ago, PinellasFB said:

What you hope for in any relative power index algorithm is it is generally correct and then grows more accurate as the season progresses, resolving inevitable anomalies over time.  An anomaly is one team ranked over another team that is clearly inferior to the team it is ranked ahead of.  Looking at the FHSAA-maxpreps rankings, I think this is generally true?  There are still anomalies but they are getting corrected each week.   No ranking will be perfect or even near perfect but you need one that is at least good enough.  Only time will tell if the fhsaa bastardization of the maxpreps ranking is good enough.  It's clearly inferior to the standard maxpreps rankings though. 

Perhaps instead of margin of victory, they can implement margin of defeat?   A win by 1 point or 70 counts the same, but a loss by 1 is much better than a loss by 70.  At least we can reward teams for playing a tough schedule with a bunch of close losses.  

Who determines what a tough schedule is and how is that rated?  Opinion is one thing.  How is it validated statistically?

[PinellasFB]

5 minutes ago, Nulli Secundus said:

Who determines what a tough schedule is and how is that rated?  Opinion is one thing.  How is it validated statistically?

Statistically it's done via large time series analyses involving cross-correlation matrices, weights and regressive moving averages.  In english, this simply means you are measured how well on average you perform versus a team compared to how others performed against that same opponent. 

A simple model example:  If you scored higher than the average versus team A and then gave up less than the average versus team A, then you have performed above average with respect to all other teams who played team A.  Now if team A performs above average versus other teams as well, this will in turn improve your ranking even further.  Below average performance works the opposite way.  Once more and more teams play, the more two teams are relatable to each other since more degrees of common opponents happen, thus the model converges to optimum accuracy late in the season, smoothing over wonky early season results.

Now take away scoring margin from the above model and you can see how you lose precision in measuring how well above or below you performed against team A since now it is either W or L with no shades of gray between.

[181pl]

going back to some of Dr. D‘s work, which was really good.

I’d probably shoot for 12 districts with 8 teams per and have the top 2 qualify. FL is committed to a 5 round playoff so we’d have the last 8 qualify through a wildcard based on record against teams in their region. Could even do something like a Kansas tiebreaker between the teams that don’t automatically qualify. That would be really cool. Would need a few more teams in the large classes. But right now we have nine classes and a bunch of independents. We don’t need nine classes. We probably need six classes. The smaller classes could operate on the region model and have a four round playoff. The larger classes, 4, 5, and 6, would have five rounds of playoffs and there should be plenty of teams to fill those districts up.

[Nulli Secundus]

7 minutes ago, PinellasFB said:

Statistically it's done via large time series analyses involving cross-correlation matrices, weights and regressive moving averages.  In english, this simply means you are measured how well on average you perform versus a team compared to how others performed against that same opponent. 

A simple model example:  If you scored higher than the average versus team A and then gave up less than the average versus team A, then you have performed above average with respect to all other teams who played team A.  Now if team A performs above average versus other teams as well, this will in turn improve your ranking even further.  Below average performance works the opposite way.  Once more and more teams play, the more two teams are relatable to each other since more degrees of common opponents happen, thus the model converges to optimum accuracy late in the season, smoothing over wonky early season results.

Now take away scoring margin from the above model and you can see how you lose precision in measuring how well above or below you performed against team A since now it is either W or L with no shades of gray between.

That sounds like a fun project for next year, lol.  I might have to work that out.