Phase Two of Prospect Modeling takes us to College Productivity, (read about Phase One's Combine Score here). While a prospect's athletic measurables shows us his raw developmental tools , his statistical productivity in college shows us what he was able to do with those tools. When predicting the future statistical prowess of a prospect, logic follows that past statistical prowess would have inherent value even in a different environment. The College Productivity Model works to capitalize on this simple premise: good NFL players were once great college players.
Pitt standout, Larry Fitzgerald, sits atop the WR Productivity Model.
Getting an everlasting stud as the headliner can't be a bad thing.
As with the Combine model, identifying which stats carry the highest correlation to fantasy production in the NFL is the first step to finding usable building blocks. The tables below show the correlation of various college stats as they relate to a player's NFL fantasy Points Per Game from age 27 and under. The metrics chosen for the College Productivity Score are highlighted in yellow.
In order to further increase the value of these building block metrics, the model adjusts using two key external values: Age and Conference.
When modifying for Age and Conference, we first consider the effect on a player's collegiate output. In conferences like the Big 12 or the Pac 12, passing games flourish. Defense is a foreign concept for these schools, and the production of offensive players in these conferences is consequently inflated. Similarly, younger players often sit behind older players for reps, so a player that standouts at a young age is more valuable.
Secondly, actual NFL success (PPG) varies by collegiate conference and breakout age. We can capture the full impact of Conference and Age by combining these perspectives. Below is an example of the Age and Conference modification factors implemented in the wide receiver model. These factors are built from NFL fantasy results from 2003-2019.
With WRs, we see the heavy impact of conference when comparing Pac-12 prospects against SEC players, for example. Historically, Pac-12 players run wild, only to struggle as soon as they enter the league. Consider the long line of whiffs out of the powerhouse USC program or try to name a legit Hall of Fame WR that came out the Pac-12 in the last 20 years. Then flip over to the SEC, where the Julio Jones and AJ Greens of the world fail to put up big numbers, yet absolutely dominate at the pro level. Modifying for these values is incredibly useful. As a side note, the average NFL PPG for WR prospects who breakout in their senior season is the same as those who never cross the breakout threshold. Worth mentioning for those stuck on killing prospects who "never dominated in college."
The charts below display exactly what Age and Conference adjusting can do, raising the correlation of a player's Average Collegiate Receiving Yards from 0.396 to 0.542 and nearly doubling the R^2 value. On its own, this adjustment is more valuable for predicting fantasy success than the NFL Draft (-0.502 correlation).
Team Adjusted Dominator Over Average
Another popular component of evaluating a player's college productivity is the family of market share metrics. Breakout Age (used for age adjustment in this model) is, itself, a derivative of Dominator Rating - the comprehensive capture of a player's market share receptions, yards and touchdowns.
Dominator Rating is great. Market share can give more reliable information than raw stats and also adjusts for poor QB play or low offensive efficiency. Brilliant spreadsheet overlord, Peter Howard, invented a fantastic age adjustment of Dominator Rating with his Dominator Over Average, which compares the Dominator stats of certain ages against those of successful NFL players. I aim to take this modification even further with the creation of Team Adjusted Dominator. As seen with conference adjusting, evaluating the environment where a player produces makes a significant difference. Since Dominator focuses on the productivity ownership that a player has over his own teammates, it follows that the talent of the teammates he is dominating matters as well. Corey Davis standing out among future real estate brokers at Western Michigan is not the same as Devonta Smith shining on a team with multiple first round WRs at Alabama.
There are many fascinating methods of applying this principle, but the first iteration of this model starts with the simplest. The model measures the strength of a player's teammates using a college team's end of season rank. Dominating on a National Championship team is more valuable than winning the starting job at Malone University. The charts below illustrate the organizational advantage that Team Adjusted Dominator Over Average has over Average Dominator. Fantasy correlation of Average Dominator sits at 0.398, Peter Howard's Dominator Over Average takes it to 0.435, and Team Adjusted Dominator Over Average boosts the value above Draft Capital, taking the correlation to 0.511.
The combination of Team Adjusted Dominator Over Average and Conference Modified Age Adjusted Yards acts as the WR Production Score (Phase Two) with a stellar correlation of 0.566. When combined with Phase One's Combine Score, the model's final Pre-Draft Score finishes with a correlation of 0.609. The charts below illustrate the value of supplementing College Productivity with a fantastic athleticism metric.
Proper contextualization of college productivity and athleticism produces a model that outperforms NFL war rooms.
More details on the final Pre-Draft score are on the way, but let's review Phase Two's Production Score (sans Combine Score) for all positions:
Wide Receiver Production Model
Dominator Over Average
NFL PPG Correlation: 0.566
Running Back Production Model
Yards Per Carry
Max/Average Receiving Yards
Age Adjusted PPG
Conference Adjusted PPG
NFL PPG Correlation: 0.480
Tight End Production Model
Average Receiving Yards
Age Adjusted PPG
Conference Adjusted PPG
NFL PPG Correlation: 0.537
The Top Producers of the 2020 Rookie Class
Creating a model that's better than the NFL Draft? Sure whatever, take me to the rookies! You got it. Here are the top scorers of the College Productivity Model:
Wide Receivers - Production Score Percentile
1. Bryan Edwards, South Carolina - 96%
2. KJ Hamler, Penn State - 89%
3. Jerry Jeudy, Alabama - 89%
4. Tyler Johnson, Minnesota - 87%
5. Justin Jefferson, LSU - 85%
Running Backs - Production Score Percentile
1. Jonathan Taylor, Wisconsin - 98%
2. JK Dobbins, Ohio State - 88%
3. AJ Dillon, Boston College - 70%
4. Zack Moss, Utah - 68%
5. Cam Akers, Florida State - 64%
Tight Ends - Production Score Percentile
1. Albert Okwuegbunam, Missouri - 98%
2. Adam Trautman, Dayton - 92%
3. Brycen Hopkins, Purdue - 79%
4. Charlie Taumoepeau, Portland State - 79%
5. Hunter Bryant, Washington - 79%
Check out the entire Prospect Model Database from 2003-2020 here.