05-31-2022, 08:05 PM
(This post was last modified: 06-01-2022, 07:18 AM by Suggs. Edited 2 times in total.)
Hey everyone! I was really impressed by @br0_0ker's Draft Primer, so I wanted to make my own Mock Draft/Big Board based on a purely statistical analysis of player's performance on the DSFL and how that could translate to the ISFL. Whilst making it, however, I came a across a very interesting fact that I had brought up to other members while making past projects that I just couldn't believe: TPE does not have a strong correlation to team performance. Well is it true? Maybe yes. Maybe no. I'll explain why.
To start off, I wanted have to make a Big Board that could be referenced throughout the article and guide my Mock Draft logic, as well as a system that would guide teams as to what their priorities should be in roster construction and, as a consequence, drafting. For that, I established some steps to my project:
To establish positional values, I wanted to have a less time-consuming process than the Wins Above Average formula that I had done for the ISFL in S29, so I decided to do the only other thing that demonstrated with numbers how good a player was besides pure performance stats: TPE. To do so, I decided to make a spreadsheet with the average TPE of every position for every team and see how much each position would correlate to a team's offensive/defensive performance. So how do you determine offensive/defensive performance? Do you use yards gained/conceded or points gained/conceded? It doesn't matter, since yards gained have a strong correlation with points gained and the same thing happens for yards and points conceded, right? Well, not really.
Through a first look, it seems like there is a clear, linear relation between points scored/allowed with yards gained/allowed, right? Not quite.
The graphs above show that there isn’t a strong correlation between yards and points in neither side of the ball (although it might look good, a R-squared of 0.42 only explains 17% of the variance in the model and a R-squared of 0.6576 only explains about 41% of the variance in the model). So I tried to create a correlation between QB, WR, and RB TPE and offensive yards and points as well as LB, and CB TPE with defensive yards and points. The results were unexpected, to say the least.
While it can be noted that there are multiple LBs on every team that have different capabilities, affect the game in different ways, and the possibility of diminishing returns in LB performance and a non-linear increase in performance based in TPE increase in certain “mid-level” (from lets say, about 350 to 700), the first two concepts are not true for QBs, which also experience very low levels of correlation with performance, especially considering they are widely viewed (with reason) the two most important positions in their respective sides of the ball.
In conclusion: does that mean that TPE is not important? No, absolutely not. I just believe that the relationship between TPE is not entirely linear and simple as most would (again, with reason) imagine. What was proposed by some users that I discussed this issue with (shoutout to XaveValor for this concept. I tried to tag him but the forum wont allow me for a reason) was that since the rate in which your player increases and TPE increases is not linear (as upgrading follows a non-linear scale; you don’t always upgrade 1 of your player’s stat by the cost of 1 TPE) is completely valid and should be explored upon in the future. I might even do it eventually.
Does that mean I’m not doing a mock draft then? No, I do plan on doing one and instead measuring player performances by adapting the Wins Above Replacement metric that I developed way back in S29 to complete steps 2 and 3.
Thanks for reading!
To start off, I wanted have to make a Big Board that could be referenced throughout the article and guide my Mock Draft logic, as well as a system that would guide teams as to what their priorities should be in roster construction and, as a consequence, drafting. For that, I established some steps to my project:
- Establish positional values to determine what are the most valuable positions
- Gather stats on player's DSFL performance
- Determine how valuable individual players would be for any set ISFL team
- Understand general league positional needs
- Make Big Board based on general league positional needs and player performance
- Understand individual team positional needs
- Make Mock Draft
- ?????
- Profit
To establish positional values, I wanted to have a less time-consuming process than the Wins Above Average formula that I had done for the ISFL in S29, so I decided to do the only other thing that demonstrated with numbers how good a player was besides pure performance stats: TPE. To do so, I decided to make a spreadsheet with the average TPE of every position for every team and see how much each position would correlate to a team's offensive/defensive performance. So how do you determine offensive/defensive performance? Do you use yards gained/conceded or points gained/conceded? It doesn't matter, since yards gained have a strong correlation with points gained and the same thing happens for yards and points conceded, right? Well, not really.
Through a first look, it seems like there is a clear, linear relation between points scored/allowed with yards gained/allowed, right? Not quite.
The graphs above show that there isn’t a strong correlation between yards and points in neither side of the ball (although it might look good, a R-squared of 0.42 only explains 17% of the variance in the model and a R-squared of 0.6576 only explains about 41% of the variance in the model). So I tried to create a correlation between QB, WR, and RB TPE and offensive yards and points as well as LB, and CB TPE with defensive yards and points. The results were unexpected, to say the least.
While it can be noted that there are multiple LBs on every team that have different capabilities, affect the game in different ways, and the possibility of diminishing returns in LB performance and a non-linear increase in performance based in TPE increase in certain “mid-level” (from lets say, about 350 to 700), the first two concepts are not true for QBs, which also experience very low levels of correlation with performance, especially considering they are widely viewed (with reason) the two most important positions in their respective sides of the ball.
In conclusion: does that mean that TPE is not important? No, absolutely not. I just believe that the relationship between TPE is not entirely linear and simple as most would (again, with reason) imagine. What was proposed by some users that I discussed this issue with (shoutout to XaveValor for this concept. I tried to tag him but the forum wont allow me for a reason) was that since the rate in which your player increases and TPE increases is not linear (as upgrading follows a non-linear scale; you don’t always upgrade 1 of your player’s stat by the cost of 1 TPE) is completely valid and should be explored upon in the future. I might even do it eventually.
Does that mean I’m not doing a mock draft then? No, I do plan on doing one and instead measuring player performances by adapting the Wins Above Replacement metric that I developed way back in S29 to complete steps 2 and 3.
Thanks for reading!