Ok nerd, what's all this mean...?


Draft Analytics


Definitions

You will see the above table on the Draft page and the data there is a basis for so many of the other pages. So what does it all mean and how should you think about it?


Career Value (AV):

Approximate Value (AV) was created by PFR's founder Doug Drinen and is a great measure of player value that can be used across any position. At it's core, AV is based on how effective offensive players are at generating points or defensive players are at preventing scoring. Many may be familiar with Expected Points Added (EPA) which is great but EPA is more useful for individual play decision-making, valuing specific positons only (QB, RB, WR, TE), and measuring large macro trends (seasons, play types, etc.). AV is highly correlated with EPA as both are fundamentally based on points, but AV is created for all positions, allowing valuation of players across positions.
The actual AV number a player has will not mean anything by itself, but to put some context on the actual AV number, here are some examples and player classifications:

Issues or considerations with AV:

More detail can be found on PFR's website here: https://www.sports-reference.com/blog/approximate-value/


AV per Year:

Since AV accumulates each season, players that have been in the league longer will have higher AVs and, therefore, more total value. AV per year provides a way to compare players across draft years by measuring their average impact per year.


Position Value %:

To make the valuation number more useful, each player's value is calculated into a percentile rank based on their AV per year compared against all other players that play their position. For example, Landon Dickerson's meaningless 6.4 AV translates to a 93rd percentile, meaning he has provided more value than 93% of other interior OL. To show the benefit of calculating percentiles by position, Jalen Hurts has a career AV per year of 11.3, almost double Landon's, but Hurts' value percentile is basically the same at an elite 92%.


Class:

Groups players by their Player Value % into classifications. "Elite players" are players with a value percentile of 85% or above (90% for players in the league 3 years or less to eliminate some of the small sample size noise), "Above Average" players are next and go down to 60% percentile values, "League Average" are between 40-60%, and "Poor" are below 40% percentile players.


Value vs Expected:

An expected value has been calculated for each pick slot using historical draft data. For example, the 1st overall pick has an expected 79% player percentile based on historical draft values. The 2023 top pick, Bryce Young, has a career valuation percentile of 70% but that is 10% below expected value and 21% behind 2nd overall pick C.J. Stroud. Puka Nacua, taken at pick 177 had an expected value Where AV and AV per year give a measure of a player's actual value regardless of draft spot, Value vs. Expected measures draft efficiency or productivity and is a great view on how well teams or GMs have drafted.


Position # Drafted:

Ranks players by their position order drafted. The first CB taken will be CB1 or position #1, the second CB taken would be CB2 or position #2, and so on.


Position Value Rank:

Ranks the players by the value they generated against all others drafted at their position. Used with Position # Drafted above this shows what the draft order was vs. should have been.
Example:In the awful 2022 QB class, Brock Burdy was the 9th QB taken (Position # Drafted) but 1st in QB value (Position Value Rank).


RAS:

A player's aggregate Relative Athletic Score that gives a view on a player's athletic profile from the great site ras.football.




Improvements to the valuation model


I mentioned some of the issues or considerations with AV above and I have made several improvements to the valuation model over time to get better player value percentiles:


Reducing QB Overvaluation:

To illustrate the issue with QB overvaluation, here are some player comparisons using player value percentiles based on PFR's raw AV:

For the math curious, I used a log transformation to decrease the natural QB overvaluation. With this, Daniel Jones drops from 85% to 68% (the same as Jermaine Johnson and Chamarri Connor), Trubisky from 72% to 53% (same as Jerome Ford and Malcolm Rodriguez), and Mariota from 90% to 65% (same as Renardo Green and Cam Hart). But the actually good QBs remain properly valued - Mahomes at 100%, Lamar Jackson at 93%, Josh Allen 92%,and Jalen Hurts 92%.


Adjusting Positional Biases:

AV makes assumptions in its calculation on how to split value up across positions and works well, but has one issue that shows up with modern defensive alignments. Without getting too deep on the calculations, PFR assumes 3.6 players are on the field in the secondary which doesn't match with nickel effectively being the new base defense and overvalues LBs at the expense of CBs. A snap-adjustment is made that properly values CBs.


Adjusting for Seasonal Shifts:

AV has a clear upward bias from 2018 on which overvalued recent players. Part of this is the addition of the 17th regular season game in 2021 as more games equals more value, but that doesn't totally explain it. A moving average-based adjustment to the model is added which smooths these years out but also preserves differentiation of good draft classes vs. not good classes.