ELO ratings - Advil and Glenlivet ready

[golfaddict1]

Old formula had a flaw in it, so I amended it but still not to what I desire... any programmers in the house?   

34 games added through weekend of August 18-20  (will add and subtract games as teams rise and fall as more is not necessarily better... more games is most important)

https://docs.google.com/spreadsheets/d/10NnFEfGyyY8hyQfuaJThZLIHcqktSP3jz1Bhs-JEWn4/edit?usp=sharing

As all teams begin at 1500, obviously even 3 games in to compare will be scary... but let's see how it goes.  287 schools added so far (some don't belong and likely won't play 3 minimum games with this group of schools, but they are tied to schools to compare to with more games to use... so some states might look more realistic than others with more connecting data).    

* Pending undetermined adjustment for sos.   Also, home field advantage  ( My guess around 2.5 to 3 pts )could be added if someone knows how to do so I'll add the data needed.   

 

 

 

 

[ORabidOne]

6 minutes ago, golfaddict1 said:

If I’m going to add schools and try to attempt this nationally, I figured try the Trinity League with various K factor numbers used to compare.  The K number basically defines the ratings points value on a single game.  For a glance at how they adjust I have added 4 options and they move up or down based on only the result of the teams playing.  

Other games with comparable opponents don’t matter.  Each game a team plays only that team and it’s opponent will adjust accordingly based on that one game only and have a latest weekly rating based on prior performances plus the last one entered.  

Have a look at game pts movement using each K rating example and let me know which one looks best to use.  The goal is to not under or over emphasize a one game result, but over time it works itself out to setup a rating as shown here for the last 4 year period.  

Record and every game result for 14-17 in the Trinity League loaded.  

Thanks for any input.  

 

elo.xlsx

It's hard to get a handle on Trinity, until the final transfers are in place!

[golfaddict1]

Next step is defining every school state by state that can show enough sos games to justify adding them.  For a 4 year period I’d like to have at least 3 games a season minimum or 12 games total over a 4 year period where whatever criteria I setup doesn’t dilute the rating system too much but also doesn’t have me adding scores at 3 am either.   Thinking of using a combo of Calpreps top 250 dynasty last 4 years and Fisher’s top 225 going into 2018 as a list to start with and then if anyone would like to see a school added or removed I can edit.  

Suggesions/comments welcome.  

[Sammyswordsman]

I like it..... But what exactly are you tryna do????

[golfaddict1]

1 hour ago, Sammyswordsman said:

I like it..... But what exactly are you tryna do????

It's a chess based rating system and it's also being used in FIFA world rankings soon 

https://www.sportskeeda.com/football/fifa-revises-world-ranking-system-to-follow-a-modified-version-of-the-elo-system-now

NCAA football example from last year.   http://warrennolan.com/football/2017/elochess

 

I've been curious about how an ELO rating system could work for hsfb, so figured what the heck... problem is it's weak on connect the dots, so the criteria I need to set up to use only certain schools and games to load will need to be tinkered with and the more games of substance added the better the ratings will look (if the criteria wasn't F'd up too badly ).  If I dilute the games too much with ratings some schools will be rated unfairly too high.   

It is possible with a lot of help to load not just the elite schools but 14,000+,... the formula is in the excel sheet, the problem is loading all the schools and entering all the scores.  I can do a lot but help lol.  

[golfaddict1]

States initially excluded from the list at first glance  (18):

Alaska, Connecticut,  Delaware, Iowa, Kansas, Maine, Mass, Minn, Neb, NH, North Dakota, Oregon, Rhode Island, South Dakota, Vermont, West Virginia, Wisconsin and Wyoming   

[Ararar]

Are you going to use status of the game in your calculations? If so how are you going to determine that in OOS games?

[Belly Bob]

This isn't my thing, so I want to apologize at the front end for asking such flat-footed question, but wouldn't this track the quality of programs and not of teams? 

It would tell us (say) whether MD as a HSFB program is trending up or down over a number of seasons, but not whether MD's 2017 is better or worse than SJB's 2017 team.

Your ratings seem to have SJB rated higher than MD in 2017, both at the beginning of the season and at the end. 

By my understanding of ELO ratings, that would mean that we should've expected SJB to beat MD in an imaginary 3rd game in 2017, even after MD beat SJB twice in 2017, and once by a score of 49-24. 

So I guess I'm wondering how the ratings are supposed to tell us how good a team is at the end of this year (relative to other teams this year) if the ratings for this year are determined in large part by the ratings from previous years.

And if that's not how it's supposed to work, then how do you determine ELO ratings for week 0 in any given year?

[golfaddict1]

15 hours ago, Ararar said:

Are you going to use status of the game in your calculations? If so how are you going to determine that in OOS games?

Not sure what you mean by status of the game... strength of opponent or emphasis on a major matchup?  Can you explain further, thanks.  

OOS is of course a concern as is getting in enough competitive games to make mostly non connecting data look like it has some.    Thanks for chiming in and @Belly Bobas well.   I welcome feedback.  

[Ararar]

24 minutes ago, golfaddict1 said:

Not sure what you mean by status of the game... strength of opponent or emphasis on a major matchup?  Can you explain further, thanks.  

OOS is of course a concern as is getting in enough competitive games to make mostly non connecting data look like it has some.    Thanks for chiming in and @Belly Bobas well.   I welcome feedback.  

Yes, emphasis on major matchups.How will you determine the strength of matchups 

[golfaddict1]

10 hours ago, Belly Bob said:

This isn't my thing, so I want to apologize at the front end for asking such flat-footed question, but wouldn't this track the quality of programs and not of teams? 

It would tell us (say) whether MD as a HSFB program is trending up or down over a number of seasons, but not whether MD's 2017 is better or worse than SJB's 2017 team.

Your ratings seem to have SJB rated higher than MD in 2017, both at the beginning of the season and at the end. 

By my understanding of ELO ratings, that would mean that we should've expected SJB to beat MD in an imaginary 3rd game in 2017, even after MD beat SJB twice in 2017, and once by a score of 49-24. 

So I guess I'm wondering how the ratings are supposed to tell us how good a team is at the end of this year (relative to other teams this year) if the ratings for this year are determined in large part by the ratings from previous years.

And if that's not how it's supposed to work, then how do you determine ELO ratings for week 0 in any given year?

This isn't my thing, so I want to apologize at the front end for asking such flat-footed question, but wouldn't this track the quality of programs and not of teams? 

This isn't my thing either  and I'm still not sure this rating system really works for hsfb without a serious attempt at researching some more Trinity League type examples around the country and start to figure out a plan to have ratings for non connecting data mainly.   Thanks for replying.    My 4 year example was a rough draft to see how ratings move for a long enough period but not too long.  Also, it coincides with Calpreps 4 year dynasty period currently... which I am going to use the top 1000 W-L record to determine if schools have enough games to move, so figured a good base to start.    

It would tell us (say) whether MD as a HSFB program is trending up or down over a number of seasons, but not whether MD's 2017 is better or worse than SJB's 2017 team.

Correct, unless I am brave enough to start a season with all schools at 1500 and go from there.   I would need to use a high K factor to illicit enough movement in a small inventory of games.   Ultimately this could be a goal, but I'm more interested at the moment in trying to define how to rate schools from different states with little interaction.    There is no hiding from this issue as there are not enough OOS games interaction.    

Your ratings seem to have SJB rated higher than MD in 2017, both at the beginning of the season and at the end. 

I looked at 150 K factor.  2017 start of season SJB was at 1904 and finished at 1890, a drop of 14 pts in the year.  MD started at 1728 and finished at 1872, an increase of 144 points.   MD was definitely trending up but historically over that 4 year period SJB has fewer losses and still retained a small lead.  

By my understanding of ELO ratings, that would mean that we should've expected SJB to beat MD in an imaginary 3rd game in 2017, even after MD beat SJB twice in 2017, and once by a score of 49-24. 

Historical rating yes, but with head to head of two games in 2017 and upwards movement vs. downward,  I would say the trend for sure looks like MD.  To use a one season of every school starting at 1500 pts is fine for NCAA and 129 FBS schools or FIFA with more connect the dots ...  somehow I would need to keep the total schools in the system limited (with less connecting data).  But every school I add who has a competitive opponent to use a score, I have to add that opponent and their data... if they don't have enough competitive games, their rating won't be realistic and then there's the OOS issue to deal with.  

So I guess I'm wondering how the ratings are supposed to tell us how good a team is at the end of this year (relative to other teams this year) if the ratings for this year are determined in large part by the ratings from previous years.

Covered above but good question.  If I can get to the holy grail of a good sized pool of schools with enough defined competitive game criteria and perhaps seek Echs help with his research on state scaling movements with Calpreps... can attempt to make it a one year rating system.   But it may make Calpreps turbulence of movement seem mild .   With little interaction it has to matter with emphasis.  

And if that's not how it's supposed to work, then how do you determine ELO ratings for week 0 in any given year?

I'm still leaning towards historical ratings with trend of 2018 movement to define their season vs. other schools movements and ratings, like the end of year trends of points change between the Trinity League schools in the example used.   More numbers crunching and smoke coming out of my ears might help or make me decide to quit.   

[golfaddict1]

36 minutes ago, Ararar said:

Yes, emphasis on major matchups.How will you determine the strength of matchups 

The ratings of each school will define the strength of matchup.   A one year ELO rating system needs a starting rating for Dematha vs STA first game of the year.   We can start both at 1500 and one school will move up with a win and the other will move down with a loss or stay even with a tie...  then on to the next competitive game vs another opponent who may not have a competitive game prior, so they are still at 1500.  .   This is why I'm leaning historical with +/-  vs. last season used as a key indicator in the ratings system for the current year.  I'd like to start with something vs. nothing.   A school who is much better in 2018 than previous years I would think would jump quite a bit with a high K factor used.   I can actually use a K factor of one number for archived and another for one season if it makes any difference for improving the rating system.  

[Ararar]

19 minutes ago, golfaddict1 said:

The ratings of each school will define the strength of matchup.   A one year ELO rating system needs a starting rating for Dematha vs STA first game of the year.   We can start both at 1500 and one school will move up with a win and the other will move down with a loss or stay even with a tie...  then on to the next competitive game vs another opponent who may not have a competitive game prior, so they are still at 1500.  .   This is why I'm leaning historical with +/-  vs. last season used as a key indicator in the ratings system for the current year.  I'd like to start with something vs. nothing.   A school who is much better in 2018 than previous years I would think would jump quite a bit with a high K factor used.   I can actually use a K factor of one number for archived and another for one season if it makes any difference for improving the rating system.  

Thanks for the breakdown because the starting point was what I was wondering about 

[Belly Bob]

19 hours ago, golfaddict1 said:

[...]

  I can do a lot but help lol.  

If it's just data entry, I can help. 

Adding DLS's, BG's, and Cen10's outcomes against each other and against TL teams may or may not be helpful. I'll let you decide. 

[Belly Bob]

23 hours ago, golfaddict1 said:

If I’m going to add schools and try to attempt this nationally, I figured try the Trinity League with various K factor numbers used to compare.  The K number basically defines the ratings points value on a single game.  For a glance at how they adjust I have added 4 options and they move up or down based on only the result of the teams playing.  

Other games with comparable opponents don’t matter.  Each game a team plays only that team and it’s opponent will adjust accordingly based on that one game only and have a latest weekly rating based on prior performances plus the last one entered.  

Have a look at game pts movement using each K rating example and let me know which one looks best to use.  The goal is to not under or over emphasize a one game result, but over time it works itself out to setup a rating as shown here for the last 4 year period.  

Record and every game result for 14-17 in the Trinity League loaded.  

Thanks for any input.  

 

elo.xlsx

On a standard ELO, only wins and losses matter, not anything like margin of victory, right? 

Is that how you'd want your ratings to work? 

[golfaddict1]

17 minutes ago, Belly Bob said:

On a standard ELO, only wins and losses matter, not anything like margin of victory, right? 

Is that how you'd want your ratings to work? 

Margin of victory matters and is an integral part of the rating system.   

[BUFORDGAWOLVES]

12 minutes ago, golfaddict1 said:

Margin of victory matters and is an integral part of the rating system.   

Can this be put in a formula, lets say W + L /2- D, x2 X Y2 squared or some something. Values for w, l, d. x, y, and z represent wins, losses, differential, margin of victory, etc?

Seriously, I read ya'lls statements and no matter can't really make heads or tails of it. If you have a formula that may illustrate the process better visually.

Regards

BGW

 

[golfaddict1]

I added a cumulative (named overall and 2014) rating for 4 years as well as 2015, 2016 and 2017 seasons I listed separately so you can see just the year ratings.   I had the 2016 SJB-MD games in reverse order, which I edited and it adjusted both teams just a few pts differently.   This is the revised excel sheet.  

Copy of elo.xlsx

[golfaddict1]

1 hour ago, BUFORDGAWOLVES said:

Can this be put in a formula, lets say W + L /2- D, x2 X Y2 squared or some something. Values for w, l, d. x, y, and z represent wins, losses, differential, margin of victory, etc?

Seriously, I read ya'lls statements and no matter can't really make heads or tails of it. If you have a formula that may illustrate the process better visually.

Regards

BGW

 

=IF($A51=E$1,E50+$B$1*(($B51/($B51+$D51))-(E50/(E50+LOOKUP($C51,$E$1:$Q$1,$E50:$Q50)))),IF($C51=E$1,E50+$B$1*(($D51/($B51+$D51))-(E50/(E50+LOOKUP($A51,$E$1:$Q$1,$E50:$Q50)))),E50))   

1 to 50 are the rows,  B1 is the K factor that adjusts game result and mov rating emphasis... can make games worth more or less (I prefer K factor of 150 at the moment).  Scores and mov calculated against both opponents current ratings at game time and spits out a new rating for those two teams only.  Anything more in depth I can't help... I'm just piggybacking on the ELO formula to the best of my knowledge.  

[golfaddict1]

 

[BUFORDGAWOLVES]

4 hours ago, golfaddict1 said:

=IF($A51=E$1,E50+$B$1*(($B51/($B51+$D51))-(E50/(E50+LOOKUP($C51,$E$1:$Q$1,$E50:$Q50)))),IF($C51=E$1,E50+$B$1*(($D51/($B51+$D51))-(E50/(E50+LOOKUP($A51,$E$1:$Q$1,$E50:$Q50)))),E50))   

1 to 50 are the rows,  B1 is the K factor that adjusts game result and mov rating emphasis... can make games worth more or less (I prefer K factor of 150 at the moment).  Scores and mov calculated against both opponents current ratings at game time and spits out a new rating for those two teams only.  Anything more in depth I can't help... I'm just piggybacking on the ELO formula to the best of my knowledge.  

K... thanks, I think, just kidding.

Little more involved than a simple algebraic formula.

Keep on, keeping on.

Regards,

BGW

[Belly Bob]

Any updates?

[Belly Bob]

On 7/13/2018 at 7:32 PM, BUFORDGAWOLVES said:

K... thanks, I think, just kidding.

[...]

I appreciate the pun. 

[golfaddict1]

On 7/15/2018 at 12:43 PM, Belly Bob said:

Any updates?

Sun was out all weekend, Wimbledon, Word Cup and Track and Field events watched... no updates yet.  Will try to create smoke from my ears this week, time permitting.  

 

[Belly Bob]

1 minute ago, golfaddict1 said:

Sun was out all weekend, Wimbledon, Word Cup and Track and Field events watched... no updates yet.  Will try to create smoke from my ears this week, time permitting.  

 

PM me if you want me to help in some way. I've been reading a bit on rating systems since you made this thread. I have the Glicko-2 system paper on my desk right now, which I got from Professor Glickman's page. That doesn't count for anything, I know, but the point is that I've got the thing on my mind and I think it would be fun to work on a rating system, even if it's not likely to work very well on a national scale, and even if I, myself, won't be able to contribute much.