Coach55's Parity Calculator - Predictive index/Power Rankings

[Coach55]

If you saw my previous post about the Eagles soft schedule, I was frustrated with the scheduling disparity of teams and determined looking at a team's record is a flawed way of approaching how good a team is.  Therefore I built a model in an attempt to determine how good teams are based on how well they play their opponents - entirely dictated based on the scores of the games.  The basic thought process is if BUF beats NE by 7 and NE beats MIA by 3, then BUF should beat MIA by 10.  Thus a 1 point loss to a really good team could be worth more than a 10 point win against a garbage team.  By applying that process to EVERY game played for the season it creates a ranking of teams as well as a predictive index of how much a team should win by if they play another team.  As more games are played, the statistics should become more accurate.  

 

A few notes on my calculations - the home team gets an automatic 3 points for playing at home and score differentials are capped at 21 (thus Buffalo beating Pittsburgh 38-3 is equivalent to Buffalo winning 24-3).  The model is all indexed by the worst team in the league, who is given an index score of 50 (who is currently Carolina).  Note that the index numbers don't adjust for injuries, it is entirely statistical

 

Based on this Predictive Index, the top 10 teams in the NFL are as follows:

Buffalo 74.46

Baltimore 69.68

Cincinnati 67.87

Philadelphia 66.38

Kansas City 65.08

Miami 64.84

San Fran 63.04

Dallas 62.65

Jacksonville 62.62

Tampa Bay 62.56

 

Thus - if the Bills are to play KC this week in KC, I would expect the Bills to Beat KC by 6.5 (74.46 - 65.08 = 9.39 - 3 for being on the road = 6.39, rounded to 6.5).  The current spread is KC +3, so I would be taking the Bills this week.  

 

In order to test this out, below for this week are my picks based on the model vs. the spread, along with my expected margin of victory.   

CHI -6, ATL +4.5, NE 0, NYJ 0, JAX -4, MIA -9, CIN -12, BAL -8.5, PIT +1.5, CAR +7.5, SEA +1, BUF -6.5, PHI -6.5, LAC -7

 

As this doesn't account for injuries, Miami showing a big skew.  My bet of the week is Cincinnati big (by 12) over New Orleans, who is favored by 2. 

 

Let's see how this plays out. 

 

 

 

[Marcus Aurelius]

Be sure to post the results of this algorithm of yours at the end of the regular season.

[Big Turk]

This could be very interesting betting wise if it can model anything above 60% of games.

[Einstein]

2 minutes ago, Big Turk said:

This could be very interesting betting wise if it can model anything above 60% of games.

 

Not possible.

[GunnerBill]

So DVOA by Football Outsiders already tries to account for relative strength of performance and opponent. 

 

Their current top 10:

 

1. Buffalo

2. Philadelphia

3. Baltimore

4. Tampa Bay

5. San Francisco

6. Dallas

7. Jacksonville

8. Kansas City

9. Green Bay

10. Seattle

 

Now DVOA normally say it is after week 6 that you really have enough data for it to start to become meaningful. But that is their list.

 

 

[Shortchaz]

This is for football games played in vacuums 

[Neo]

6 minutes ago, Big Turk said:

This could be very interesting betting wise if it can model anything above 60% of games.

It would be greater than perpetual motion, or nuclear power from water, or alchemy … if it can model greater than 51%!

 

 

[Big Turk]

19 minutes ago, GunnerBill said:

So DVOA by Football Outsiders already tries to account for relative strength of performance and opponent. 

 

Their current top 10:

 

1. Buffalo

2. Philadelphia

3. Baltimore

4. Tampa Bay

5. San Francisco

6. Dallas

7. Jacksonville

8. Kansas City

9. Green Bay

10. Seattle

 

Now DVOA normally say it is after week 6 that you really have enough data for it to start to become meaningful. But that is their list.

 

 

 

Bills are the best team in the NFL and it isn't particularly close right now. Not only the best team but they are a historic team right now.  Some of the metrics they are putting up are absolutely ridiculous. They have punted 5 times in 5 games with the starters in. Are you freaking kidding me?

Edited October 11, 2022 by Big Turk

[SageAgainstTheMachine]

So this is really cool and I applaud the work. Want to say that first.

 

I think the flaw is the amount of sample size you’d need for it to be meaningful based on the variance in how points get scored and then the way rosters get churned during off seasons.  In other words you’re only getting 17 games per team when you probably need hundreds to approach reliability in forecasting anything.  And then trying to achieve that data over the course of multiple seasons becomes useless when the players change teams en masse.

[dorquemada]

I played around with some of this (it was in SPSS, and it was 30 years ago, but same general idea).  Once I got the model set up and power ranking the teams they way I thought it should, I went back and played with prior year stats and tried to back into a predictive value.   I will say, it was a whole bunch harder then because there wasn't a million websites with stats to copy/paste. For that matter, the web itself only barely existed though I did find a FTP repository with historical NFL stats.  The net was a formula that could most weeks come in +2 games for the week which was enough to keep me in beer and skittles and confound my bookie.  Oh, Ian, you never saw it coming!

 

Anyway, my suggestion is run last year's stats into your model and see how it does.

[Coach55]

2 minutes ago, SageAgainstTheMachine said:

So this is really cool and I applaud the work. Want to say that first.

 

I think the flaw is the amount of sample size you’d need for it to be meaningful based on the variance in how points get scored and then the way rosters get churned during off seasons.  In other words you’re only getting 17 games per team when you probably need hundreds to approach reliability in forecasting anything.  And then trying to achieve that data over the course of multiple seasons becomes useless when the players change teams en masse.

I realize that the sample size is small as currently each team only has 5 games, however, the database currently has 80 games.  By week 10, we will be double that.  Although not deemed statistically significant, it should provide enough data to generate relatively meaningful statistics.  I would say that anything less than 5 games would not have much accuracy.  From games 6-10, I would expect it to improve and then once it hits 10, it will probably level off (but this is just an educated guess).  Unfortunately given the roster churn in the offseason, using prior year's data doesn't really help either.  This is just more of a fun experiment.  

2 minutes ago, dorquemada said:

I played around with some of this (it was in SPSS, and it was 30 years ago, but same general idea).  Once I got the model set up and power ranking the teams they way I thought it should, I went back and played with prior year stats and tried to back into a predictive value.   I will say, it was a whole bunch harder then because there wasn't a million websites with stats to copy/paste. For that matter, the web itself only barely existed though I did find a FTP repository with historical NFL stats.  The net was a formula that could most weeks come in +2 games for the week which was enough to keep me in beer and skittles and confound my bookie.  Oh, Ian, you never saw it coming!

 

Anyway, my suggestion is run last year's stats into your model and see how it does.

That is a great idea.  If I get the time, I will see how well it works.  May be a few weeks. 

[Airseven]

1. Chiefs
2. Eagles
3. Bills
4. Bucs
5. 49ers
6. Ravens
7. Chargers
8. Vikings
9. Cowboys
10. Giants
11. Packers
12. Bengals
13. Titans
14. Rams
15. Jets
16. Dolphins
17. Patriots
18. Colts
19. Jaguars
20. Cardinals
21. Saints
22. Seahawks
23. Raiders
24. Browns
25. Broncos
26. Bears
27. Falcons
28. Lions
29. Texans
30. Commanders
31. Steelers
32. Panthers

[Big Turk]

5 minutes ago, Airseven said:

1. Chiefs
2. Eagles
3. Bills
4. Bucs
5. 49ers
6. Ravens
7. Chargers
8. Vikings
9. Cowboys
10. Giants
11. Packers
12. Bengals
13. Titans
14. Rams
15. Jets
16. Dolphins
17. Patriots
18. Colts
19. Jaguars
20. Cardinals
21. Saints
22. Seahawks
23. Raiders
24. Browns
25. Broncos
26. Bears
27. Falcons
28. Lions
29. Texans
30. Commanders
31. Steelers
32. Panthers

 

Bills point differential is +91, almost double the next closest teams in the Eagles and 49ers at +47. chiefs are 4th at +34.  

 

I think the Bills will put to rest who the best team is on Sunday 

Edited October 11, 2022 by Big Turk

[Shortchaz]

1 hour ago, Airseven said:

1. Chiefs
2. Eagles
3. Bills
4. Bucs
5. 49ers
6. Ravens
7. Chargers
8. Vikings
9. Cowboys
10. Giants
11. Packers
12. Bengals
13. Titans
14. Rams
15. Jets
16. Dolphins
17. Patriots
18. Colts
19. Jaguars
20. Cardinals
21. Saints
22. Seahawks
23. Raiders
24. Browns
25. Broncos
26. Bears
27. Falcons
28. Lions
29. Texans
30. Commanders
31. Steelers
32. Panthers

This looks pretty accurate 

[SACTOBILLSFAN]

5 hours ago, Big Turk said:

This could be very interesting betting wise if it can model anything above 60% of games.

It can't. Especially with just blindly adjusting home field as 3 points and not taking into account things like wind and playcall tendencies. 

[Chaos]

6 hours ago, Coach55 said:

If you saw my previous post about the Eagles soft schedule, I was frustrated with the scheduling disparity of teams and determined looking at a team's record is a flawed way of approaching how good a team is.  Therefore I built a model in an attempt to determine how good teams are based on how well they play their opponents - entirely dictated based on the scores of the games.  The basic thought process is if BUF beats NE by 7 and NE beats MIA by 3, then BUF should beat MIA by 10.  Thus a 1 point loss to a really good team could be worth more than a 10 point win against a garbage team.  By applying that process to EVERY game played for the season it creates a ranking of teams as well as a predictive index of how much a team should win by if they play another team.  As more games are played, the statistics should become more accurate.  

 

A few notes on my calculations - the home team gets an automatic 3 points for playing at home and score differentials are capped at 21 (thus Buffalo beating Pittsburgh 38-3 is equivalent to Buffalo winning 24-3).  The model is all indexed by the worst team in the league, who is given an index score of 50 (who is currently Carolina).  Note that the index numbers don't adjust for injuries, it is entirely statistical

 

Based on this Predictive Index, the top 10 teams in the NFL are as follows:

Buffalo 74.46

Baltimore 69.68

Cincinnati 67.87

Philadelphia 66.38

Kansas City 65.08

Miami 64.84

San Fran 63.04

Dallas 62.65

Jacksonville 62.62

Tampa Bay 62.56

 

Thus - if the Bills are to play KC this week in KC, I would expect the Bills to Beat KC by 6.5 (74.46 - 65.08 = 9.39 - 3 for being on the road = 6.39, rounded to 6.5).  The current spread is KC +3, so I would be taking the Bills this week.  

 

In order to test this out, below for this week are my picks based on the model vs. the spread, along with my expected margin of victory.   

CHI -6, ATL +4.5, NE 0, NYJ 0, JAX -4, MIA -9, CIN -12, BAL -8.5, PIT +1.5, CAR +7.5, SEA +1, BUF -6.5, PHI -6.5, LAC -7

 

As this doesn't account for injuries, Miami showing a big skew.  My bet of the week is Cincinnati big (by 12) over New Orleans, who is favored by 2. 

 

Let's see how this plays out. 

 

 

 

Over the course of seasons this will not yield useful results. It has been done before. (Note it did work very well in 1982 indicating betting on the Redskins to cover virtually every week. This did provide good spending money for a semester in college.  Sadly this did not work out permanently as a system. 

https://www.statmuse.com/nfl/ask/1982-washington-redskins-against-the-spread  10-2.  One of my best runs ever. 
 

Edited October 12, 2022 by Chaos

[MiltonWaddams]

OK, so I put $10,000 down according to your algorithm. Please don’t tell my wife.

[MJS]

Nice work. Seems interesting.

 

I guess I would want to see other stats incorporated. The score is one thing, but it doesn't tell the whole story. 3rd down proficiency would be a big one. Explosive plays. Lots of other stats.

[HardyBoy]

15 hours ago, Shortchaz said:

This is for football games played in vacuums 

 

Bills Pats last year was definitely played in a vacuum (cleaner).

[Sammy Watkins' Rib]

On 10/11/2022 at 11:24 AM, Coach55 said:

 

 

In order to test this out, below for this week are my picks based on the model vs. the spread, along with my expected margin of victory.   

CHI -6, ATL +4.5, NE 0, NYJ 0, JAX -4, MIA -9, CIN -12, BAL -8.5, PIT +1.5, CAR +7.5, SEA +1, BUF -6.5, PHI -6.5, LAC -7

 

As this doesn't account for injuries, Miami showing a big skew.  My bet of the week is Cincinnati big (by 12) over New Orleans, who is favored by 2. 

 

Let's see how this plays out. 

 

 

 

 

I put a little action on a 10 team teaser and parlay with some of the above that I also liked. Starting with the Bears tonight. We'll see how it goes.