The goal of this is to predict NFL success as accurately as possible solely using analytics. Specifically, I’m predicting Pro Football Reference’s Approximate Value Per Game (AV/G) statistic using models that have college production, athletic testing, and consensus big board data.
After analyzing AV/G, I found that the statistic isn’t representative of positional spending and consensus positional value, so I’ve made slight adjustments in an attempt to have my top 300 big board reflect positional value, something that seems to be underrated during the draft. Adjusting for positional value will certainly make some people #MadOnline because the quarterbacks will be ranked higher than you are used to and the running backs and safeties will be ranked lower than you are used to. If you have a problem with that, ignore the overall ranking and focus on where they are ranked within their position.
Another thing that needs to be made abundantly clear is that these rankings include each prospect’s projected draft pick. There is a huge difference between ranking prospects with their draft capital in mind (that’s what I’m doing in the NFL Draft Analytics Top 300) and ranking prospects as if they were all drafted with the same overall pick. For example, projected top-15 pick Daniel Jones and projected Day 2 pick Will Grier basically have the same projection, but if Grier and Jones were both projected to be the 10th overall pick, Grier would be ranked much, much higher than Jones.
Lastly, one improvement I’ve had in modeling for the NFL Draft is to separate positions into different archetypes. Like Josh Norris has noted on the Rotoworld Football Podcast -- subscribe and leave a review -- the weight difference between receivers Marquise Brown and D.K. Metcalf is 62 pounds, so it’d be silly to use the same model for each receiver. For that reason, I’ve split running backs, wide receivers, defensive tackles, and edge rushers into groups using weight or height as the separator. Doing so helped increase the out of sample r-squares for each model.
If you’re looking for the top 300 rankings, head over to Page 4.
QB model variables: draft capital, strength of schedule, age, completion percentage, overall passing efficiency, passing touchdown efficiency, and rushing production.
Rank - Prospect (historical percentile rank) - overall rank (projected overall pick)
QB1 - Kyler Murray (100th percentile QB) - 1st overall (projected 1st overall pick)
QB2 - Dwayne Haskins (91st percentile QB) - 2nd overall (projected 10th overall pick)
QB3 - Drew Lock (83rd percentile QB) - 4th overall (projected 11th overall pick)
QB4 - Will Grier (62nd percentile QB) - 13th overall (projected 80th overall pick)
QB5 - Daniel Jones (61st percentile QB) - 14th overall (projected 17th overall pick)
QB6 - Ryan Finley (42nd percentile) - 42nd overall (projected 136th overall pick)
QB7 - Tyree Jackson (35th percentile) - 65th overall (projected 104th overall pick)
QB8 - Clayton Thorson (28th percentile) - 96th overall (projected 148th overall pick)
QB9 - Jarrett Stidham (27th percentile) - 97th overall (projected 126th overall pick)
QB10 - Gardner Minshew (19th percentile) - 156th overall (projected 235th overall pick)
QB11 - Brett Rypien (18th percentile) - 166th overall (projected 207th overall pick)
QB12 - Jordan Ta’amu (4th percentile) - 281st overall (projected 225th overall pick)
QB13 - Easton Stick (4th percentile) - 282nd overall (projected 254th overall pick)
QB14 - Nick Fitzgerald (3rd percentile) - 288th overall (projected UDFA)
QB15 - Jake Browning (2nd percentile) - 297th overall (projected UDFA)
QB16 - Trace McSorley (1st percentile) - 314th overall (projected UDFA)
QB17 - Kyle Shurmur (1st percentile) - 325th overall (projected UDFA)
QB rankings without draft capital for those curious: 1) Kyler Murray, 2) Dwayne Haskins, 3) Will Grier, 4) Gardner Minshew, 5) Ryan Finley, 6) Easton Stick, 7) Drew Lock, 8) Nick Fitzgerald, 9) Brett Rypien, 10) Jake Browning, 11) Clayton Thorson, 12) Trace McSorley, 13) Tyree Jackson, 14) Jarrett Stidham, 15) Daniel Jones, 16) Jordan Ta'amu, and 17) Kyle Shurmur. … Underrated: Dwayne Haskins if he falls as reported, Will Grier, Gardner Minshew, Ryan Finley. … Overrated: Daniel Jones, Jarrett Stidham, Clayton Thorson, Tyree Jackson.
Everything I've built for this column uses historical data because I need to be able to test it to build models. However, there are a lot of new metrics that I believe will be quite predictive once there is enough historical data to test its predictability. The two leaders in these new-age analytics are Pro Football Focus and Sports Info Solutions. Following both of them on Twitter is a must, but I'm sure you already are.
One metric I'm extra bullish on for quarterbacks is on-target percentage, especially when adjusted for air yards. Josh Hermsmeyer had a great column for FiveThirtyEight that showed there is some predictability in adjusted completion percentage, and Sports Info Solutions' on-target percentage is quite similar. Here's how the top four quarterbacks from this class rank up, but it's important to note that Will Grier is more accurate than Lock and Jones:
Small RB (under 215 lbs.) model variables: draft capital, weight, speed, weight-adjusted speed, agility, rushing market share, rushing touchdown efficiency, and receiving production. … Big RB (at least 215 lbs.) model variables: draft capital, strength of schedule, agility, breakout age, rushing market share, rushing yardage efficiency, and receiving production.
Rank - Prospect (historical percentile rank) - overall rank (projected overall pick)
RB1 - Joshua Jacobs (86th percentile Big RB) - 56th overall (projected 30th overall pick)
RB2 - Damien Harris (74th percentile Big RB) - 84th overall (projected 85th overall pick)
RB3 - Darrell Henderson (81st percentile Small RB) - 88th overall (projected 74th overall pick)
RB4 - David Montgomery (59th percentile Big RB) - 115th overall (projected 82nd overall pick)
RB5 - Miles Sanders (61st percentile Small RB) - 121st overall (projected 77th overall pick)
RB6 - Rodney Anderson (47th percentile Big RB) - 134th overall (projected 143rd overall pick)
RB7 - Bryce Love (55th percentile Small RB) - 144th overall (projected 146th overall pick)
RB8 - Justice Hill (44th percentile Small RB) - 151st overall (projected 126th overall pick)
RB9 - Ryquell Armstead (31st percentile Big RB) - 172nd overall (projected 139th overall pick)
RB10 - Darwin Thompson (37th percentile Small RB) - 175th overall (projected 199th overall pick)
RB11 - James Williams (35th percentile Small RB) - 176th overall (projected 250th overall pick)
RB12 - Alex Barnes (26th percentile Big RB) - 187th overall (projected 254th overall pick)
RB13 - Benny Snell (25th percentile Big RB) - 188th overall (projected 254th overall pick)
RB14 - Dexter Williams (33rd percentile Small RB) - 192nd overall (projected 161st overall pick)
RB15 - Devin Singletary (32nd percentile Small RB) - 193rd overall (projected 95th overall pick)
RB16 - Trayveon Williams (26th percentile Small RB) - 199th overall (projected 135th overall pick)
RB17 - Alexander Mattison (21st percentile Big RB) - 207th overall (projected 229th overall pick)
RB18 - Mike Weber (25th percentile Small RB) - 210th overall (projected 206th overall pick)
RB19 - Myles Gaskin (15th percentile Small RB) - 226th overall (projected 189th overall pick)
RB20 - Devine Ozigbo (14th percentile Big RB) - 227th overall (projected 251st overall pick)
RB21 - Qadree Ollison (12th percentile Big RB) - 228th overall (projected 250th overall pick)
RB22 - Elijah Holyfield (10th percentile Big RB) - 245th overall (projected 254th overall pick)
RB23 - Karan Higdon (8th percentile Small RB) - 260th overall (projected 254th overall pick)
RB24 - Jalin Moore (3rd percentile Small RB) - 284th overall (projected UDFA)
RB25 - Kerrith Whyte (2nd percentile Small RB) - 291st overall (projected UDFA)
RB26 - Travis Homer (1st percentile Small RB) - 304th overall (projected UDFA)
RB27 - L.J. Scott (1st percentile Big RB) - 315th overall (projected UDFA)
RB28 - Jordan Scarlett (1st percentile Small RB) - 321st overall (projected UDFA)
These running backs are sorted by projected approximate value per game. Since there are two models here, it’s possible that a running back with a worse percentile is listed above a running back with a higher percentile. … Underrated: Darrell Henderson, Bryce Love if healthy, Darwin Thompson, James Williams, Alexander Mattison, Mike Weber. … Overrated: Josh Jacobs if he goes top-25, David Montgomery, Alex Barnes, Benny Snell.
One thing I learned: SPARQ, an athletic composite score, is really interesting, but it's a flawed metric. The SPARQ formula doesn't change from position to position, and I think we can all agree that the type of athleticism -- strength vs. long-speed vs. short-area burst -- changes from position to position. For example, the bench press doesn't matter for corners, but the SPARQ formula doesn't adjust for that, so the result for a particular player can trick us into thinking a prospect is more athletic than what they actually are. One perfect example is Kansas State RB Alex Barnes:
The good news is that I've created position-adjusted SPARQ formulas (they'll be coming out soon), so we (or at least myself) won't be tricked into thinking a prospect is more athletic than what they actually are.