Firstly, I would like to extend my sincerest apology to the grader who has to read this.
Secondly, a warning:
THIS POST CONTAINS MATHS/STATS. IF THAT'S NOT YOUR THING, JUST SKIP TO MY TRADE VALUE CHART AT THE BOTTOM AND SAVE YOURSELF A HEADACHE
And yes, I'm English, so it's "maths".
So, yesterday Beefstu409 posted his take on a Draft Pick Value chart. Nice article and an interesting idea, but I thought I had a better idea for the concept so I had to stea... ahem... use it and see what I could put together myself. Sorry Stu.
The Method
I took data from the TPE tracker and ISFL Wiki on every ISFL draftee from S21 - S23 inclusive. For every player I assigned the following two values:
1) Relative Draft Position
To compensate for different sized draft classes, I normalised draft positions on a scale of 0 to 100, where 0 is first pick and 100 is last pick, using
FORMULA A: (Pick - 1)/(Draft class size - 1) x 100
2) Player TPE Value
Basically, this is a player's TPE earnings as the percentage of the TPE of the top earner in their class, which I took to be the theoretical max. This, again, normalises across different years, and is given on a scale of 0 to 100, with a few notes:
a) All earnings are calculated -50, for obvious reasons
b) Inactive or retired players are awarded 0, as there's no value to an ISFL team in being gone within 4 seasons
c) Any players below 40% of the theoretical max earnings were awarded 0. There's very little realistic prospect of them ever contributing towards an ISFL team, and therefore 0 value
I then graphed all 506 draftees, and got this:
For a bit of context, that blue dot in the top left is max earner and 1oa Otters Safety Prince Vegeta, and that blue line at the bottom right is every draft pick that's been skipped over by annoyed casters.
What this gave me (read: Excel) was the ability to calculate the polynomial regression curve shown in red above. This gives us the expected value at any given relative draft position, and is defined by the following formula:
FORMULA B: y = 0.00000000328341289518042x^6 - 0.00000100796403723415x^5 + 0.000112720499117928x^4 - 0.00541276203068841x^3 + 0.107368643144085x^2 - 2.26615167013438^x + 91.1443592399927
That's gonna form my base for this.
Of course, this needs to be tweaked a bit. The first few picks are often amazing users, sim savants, locker room guys, all that stuff, and the first few picks have the advantage of added choice. So my final model is the following:
Picks 1-5 have a value of 5000, 3000, 2000, 1500 and 1000 respectively.
Picks 6 - ?* have a value of... the result of Formula A substituted into the right hand side of Formula B, then multiplied by 10 - bigger numbers always look cooler
Picks ?**- End have a value of 0.
...where ?* is the final pick with a Relative Draft Position below 70 and ?** is the pick after that.
Well, with all that done, here's an example of a draft value chart for a draft with 140 picks, using my method:
Anticipated questions:
Dude...why?
I get bored.
What's the R squared on that regression curve?
0.65
You didn't really need to give that long formula though did you?
Nah, but it makes me look... smarter? I dunno, if anyone wanted to steal this, it's there for them.
Could this be adapted to apply to the DSFL draft?
Well, this kinda hinges on ISFL teams drafting reasonably well to make a correlation between draft position and future performance, and ISFL teams have much better info than DSFL ones. So.... not really.
Do you really think that the last 30% or so of draft picks are just worthless?
Absolutely, as do some GMs, judging by yesterday's trades. There's good cases for and against the concept of a UDFA in this league I think.
But seriously, why would you do this?
1.5x draft media week, baby... wait, this will have like no words in it. Damn.
Secondly, a warning:
THIS POST CONTAINS MATHS/STATS. IF THAT'S NOT YOUR THING, JUST SKIP TO MY TRADE VALUE CHART AT THE BOTTOM AND SAVE YOURSELF A HEADACHE
And yes, I'm English, so it's "maths".
So, yesterday Beefstu409 posted his take on a Draft Pick Value chart. Nice article and an interesting idea, but I thought I had a better idea for the concept so I had to stea... ahem... use it and see what I could put together myself. Sorry Stu.
The Method
I took data from the TPE tracker and ISFL Wiki on every ISFL draftee from S21 - S23 inclusive. For every player I assigned the following two values:
1) Relative Draft Position
To compensate for different sized draft classes, I normalised draft positions on a scale of 0 to 100, where 0 is first pick and 100 is last pick, using
FORMULA A: (Pick - 1)/(Draft class size - 1) x 100
2) Player TPE Value
Basically, this is a player's TPE earnings as the percentage of the TPE of the top earner in their class, which I took to be the theoretical max. This, again, normalises across different years, and is given on a scale of 0 to 100, with a few notes:
a) All earnings are calculated -50, for obvious reasons
b) Inactive or retired players are awarded 0, as there's no value to an ISFL team in being gone within 4 seasons
c) Any players below 40% of the theoretical max earnings were awarded 0. There's very little realistic prospect of them ever contributing towards an ISFL team, and therefore 0 value
I then graphed all 506 draftees, and got this:
For a bit of context, that blue dot in the top left is max earner and 1oa Otters Safety Prince Vegeta, and that blue line at the bottom right is every draft pick that's been skipped over by annoyed casters.
What this gave me (read: Excel) was the ability to calculate the polynomial regression curve shown in red above. This gives us the expected value at any given relative draft position, and is defined by the following formula:
FORMULA B: y = 0.00000000328341289518042x^6 - 0.00000100796403723415x^5 + 0.000112720499117928x^4 - 0.00541276203068841x^3 + 0.107368643144085x^2 - 2.26615167013438^x + 91.1443592399927
That's gonna form my base for this.
Of course, this needs to be tweaked a bit. The first few picks are often amazing users, sim savants, locker room guys, all that stuff, and the first few picks have the advantage of added choice. So my final model is the following:
Picks 1-5 have a value of 5000, 3000, 2000, 1500 and 1000 respectively.
Picks 6 - ?* have a value of... the result of Formula A substituted into the right hand side of Formula B, then multiplied by 10 - bigger numbers always look cooler
Picks ?**- End have a value of 0.
...where ?* is the final pick with a Relative Draft Position below 70 and ?** is the pick after that.
Well, with all that done, here's an example of a draft value chart for a draft with 140 picks, using my method:
Anticipated questions:
Dude...why?
I get bored.
What's the R squared on that regression curve?
0.65
You didn't really need to give that long formula though did you?
Nah, but it makes me look... smarter? I dunno, if anyone wanted to steal this, it's there for them.
Could this be adapted to apply to the DSFL draft?
Well, this kinda hinges on ISFL teams drafting reasonably well to make a correlation between draft position and future performance, and ISFL teams have much better info than DSFL ones. So.... not really.
Do you really think that the last 30% or so of draft picks are just worthless?
Absolutely, as do some GMs, judging by yesterday's trades. There's good cases for and against the concept of a UDFA in this league I think.
But seriously, why would you do this?
1.5x draft media week, baby... wait, this will have like no words in it. Damn.