09-20-2021, 07:16 PM
This thread is going to be used to house some fun numbers projects that I am running this season. To begin, we will start with a Pythagorean projection of all NFL teams, Points over Expected (PoE) for all teams and SRS for all teams. This will probably grow into other metrics as the season progresses and my projects do as well, but that is the starting point. Please note, these numbers don't tell us a story YET. They won't begin to take much meaning until next week because our dataset is too small right now.
Pythagorean Projections - If you're a baseball fan, you'll be familiar with this. This is a formula that can be used to predict what a teams final record will be by using points for and points against. The formula looks like this...
A = Points for/Points against
(A)^2.37/(A)^2.37 + 1
To build confidence in this, we can look at last season. The 2020 Cincinnati Bengals scored 311 points and allowed 424 points. If you take 311/424, you're going to get 0.733. Now, take 0.733 and apply an exponent of 2.37. This will bring a result of 0.478. Now, take 0.478 and divide it by 1.478 and you will get 0.32. This model predicts that the 2020 Cincinnati Bengals won 32% of their games, or five games. In reality, Cincinnati won 28% of their games, or four games. The model predicts that the 2020 Oakland Raiders won 44% of their games (seven games). In reality, they won 8. You're going to find some teams that 'overperformed' or 'underperformed' by a couple of games due to the limited number of games in an NFL season. The Steelers, Chiefs and Bills are all an example of this. The model is pretty accurate at getting within a game or two of how a team should finish, usually. The Browns are the major exception. The model would predict them to be a .500 team, roughly, and they won 11 games plus a playoff victory. That's football, however. Volatility is real and 'any given Sunday' is as well. Attached is what the predictions currently look like. Keep in mind, these mean NOTHING right now. We don't have enough data to say anything with any confidence, because there has only been two games.
![[Image: f5urLmA.jpg]](https://i.imgur.com/f5urLmA.jpg)
Points over Expected (PoE) - This metric is used to try to describe how well a team is performing relative to expectations. Say it is week 10, and you have Team A vs Team B. Below are their points for and points against...
Team A || 100 points for || 140 points against
Team B || 160 points for || 120 points against
By having these numbers, we can predict a final score of 15-11 in favor of team B. The 15 points is the average of team B's points per game and team A's points allowed per game (16,14). The 11 points is the average of team A's points per game and team B's points per game allowed (10, 12). Now, if team A wins the game 20-13, then team A is going to see their offensive PoE increase by +9 and their defensive PoE decrease by -1 (negative numbers are good). So, overall, team A's PoE would increase by 10.
I am unable to show an example of what this looks like currently because you need at least three games to get the data that you want for this, but this will come out next week.
SRS - Otherwise known as Simple Rating System, this takes into account your margin of victory (MOV) and strength of schedule (SOS) to get a simple team rating. The SOS is calculated by adding up the MOV for all opponents that have been played, but not future opponents. So, this is descriptive only. Please see attached for current SRS numbers. Again, these won't start meaning much until next week where we start building more data into the formulas. You're going to notice that despite the MNF game not being played, it will show the Lions and Packers as having played two games. I did that just to get this template out on my work from home day. I wasn't worried too much about the accuracy due to the aforementioned dataset issues. This is primarily to familiarize everyone with what we are going to be working with.
![[Image: Pn0anRL.jpg]](https://i.imgur.com/Pn0anRL.jpg)
As we move forward, I will probably create some Tableau presentations of already gathered data such as yards per play, points per play, pressure percentage allowed etc.
If anyone has any questions, please let me know. Happy to discuss further.
Pythagorean Projections - If you're a baseball fan, you'll be familiar with this. This is a formula that can be used to predict what a teams final record will be by using points for and points against. The formula looks like this...
A = Points for/Points against
(A)^2.37/(A)^2.37 + 1
To build confidence in this, we can look at last season. The 2020 Cincinnati Bengals scored 311 points and allowed 424 points. If you take 311/424, you're going to get 0.733. Now, take 0.733 and apply an exponent of 2.37. This will bring a result of 0.478. Now, take 0.478 and divide it by 1.478 and you will get 0.32. This model predicts that the 2020 Cincinnati Bengals won 32% of their games, or five games. In reality, Cincinnati won 28% of their games, or four games. The model predicts that the 2020 Oakland Raiders won 44% of their games (seven games). In reality, they won 8. You're going to find some teams that 'overperformed' or 'underperformed' by a couple of games due to the limited number of games in an NFL season. The Steelers, Chiefs and Bills are all an example of this. The model is pretty accurate at getting within a game or two of how a team should finish, usually. The Browns are the major exception. The model would predict them to be a .500 team, roughly, and they won 11 games plus a playoff victory. That's football, however. Volatility is real and 'any given Sunday' is as well. Attached is what the predictions currently look like. Keep in mind, these mean NOTHING right now. We don't have enough data to say anything with any confidence, because there has only been two games.
Points over Expected (PoE) - This metric is used to try to describe how well a team is performing relative to expectations. Say it is week 10, and you have Team A vs Team B. Below are their points for and points against...
Team A || 100 points for || 140 points against
Team B || 160 points for || 120 points against
By having these numbers, we can predict a final score of 15-11 in favor of team B. The 15 points is the average of team B's points per game and team A's points allowed per game (16,14). The 11 points is the average of team A's points per game and team B's points per game allowed (10, 12). Now, if team A wins the game 20-13, then team A is going to see their offensive PoE increase by +9 and their defensive PoE decrease by -1 (negative numbers are good). So, overall, team A's PoE would increase by 10.
I am unable to show an example of what this looks like currently because you need at least three games to get the data that you want for this, but this will come out next week.
SRS - Otherwise known as Simple Rating System, this takes into account your margin of victory (MOV) and strength of schedule (SOS) to get a simple team rating. The SOS is calculated by adding up the MOV for all opponents that have been played, but not future opponents. So, this is descriptive only. Please see attached for current SRS numbers. Again, these won't start meaning much until next week where we start building more data into the formulas. You're going to notice that despite the MNF game not being played, it will show the Lions and Packers as having played two games. I did that just to get this template out on my work from home day. I wasn't worried too much about the accuracy due to the aforementioned dataset issues. This is primarily to familiarize everyone with what we are going to be working with.
As we move forward, I will probably create some Tableau presentations of already gathered data such as yards per play, points per play, pressure percentage allowed etc.
If anyone has any questions, please let me know. Happy to discuss further.
