So I wasn't able to get around to writing up all position groups in detail, but I still collected data and built models for every position group for the 2019 draft. I compiled data on most prospects, and trimmed the list down to a top 200 big board. Without further adieu, here is my 2019 analytics big board:
https://docs.google.com/spreadsheets/d/1B5vGtL6kNdtS3cchE4yGaIMLhm4UUVr2gFOh5Y2q2bU/edit?usp=sharing
The AV Proj column is the raw model output for each prospect, while the Adj Grade column is that output divided by a factor to adjust for positional value. This pushes QBs, EDGEs, and WRs up the board while RBs and safeties fall, as approximate value does not really adjust for positional value. Additionally, I included both raw and adjusted surplus value numbers for every prospect. The surplus metric will tell you if a prospect is projected to exceed their draft expectation based on their model grade. Higher than 1 raw surplus means they exceed expectations for that draft slot and lower than 1 means they are projected to fail to reach that mark. For example, Kyler Murray is projected to be drafted number 1 by the model, and he exceeds the average QB drafted number 1 overall by about 29%.
Before anyone asks, the following prospects did not have enough data to be included in the big board:
Jawaan Taylor
Yodeny Cajuste
Martez Ivey
Rodney Anderson
Bryce Love
Kendall Sheffield
Vosean Joseph
D'Andre Walker
Deionte Thompson
Ben Powers
Here are a few of my overall thoughts:
- I'm disgusted Daniel Jones is as high as he is. He was an early starter, good runner, had a pretty good strength of schedule, and didn't rely on any one receiver, but pretty terrible in the other QB stats like completion percentage and yards per attempt. He's Bortles-ish to me, but the top percentile outcome is a more mobile Matt Ryan. I guess if I was desperate for a QB just outside the top 10 I might convince myself to take him, but I'd much rather pass that buck, trade down and take Will Grier in the late 1st area.
- Running backs don't matter. The first running back on the big board is David Montgomery at 106, and I don't disagree with that. In the modern game, NFL teams should not be prioritizing the running back position.
- Ashton Dulin is a rare WR prospect, despite being from such a small school. I'm making sure my team comes away with him on Day 3 if I'm a GM.
- If I was a team picking in the top 5 area and needed an edge rusher, I would love to be able to be able to trade out, pick up some extra picks, and draft Brian Burns. Josh Allen and Nick Bosa have higher grades, but Burns is close behind and provides more surplus value than either Allen or Bosa.
- I've seen Zach Allen listed as an edge rushed most places, but I would want to move him inside. He's smaller, but had elite production and his athletic profile plays better inside. He'd be a target for me in the early 2nd.
- I fully expect the model to be wrong about Jerry Tillery. PFF is extremely high on him, and I tend to take some stock in their predictive ability. This model was lower on Chris Jones too, so I could see Tillery being somewhat similar to that situation.
After taking a look at the premier passers in this year’s class, let’s move onto the guys that actually catch those passes. For the rest of the positions, I will be using two different models to evaluate players: a random forest model and a linear model. A random forest model is basically a collection of flow charts that each end up with a number that represents a projection for a player based on whatever metrics and stats go into the model. Think of each flowchart like a game of plinko: where a prospect enters at the top, passed through each peg based on a certain criteria, and then ends up in a bin at the bottom. Using thousands of these flow charts, we can come to a number that can be used to forecast a player’s NFL production. Typically, the random forest models tend to deviate more from the NFL consensus while the linear models end up more conservative. I will list both numbers for each prospect, and their average, which I use to rank each player. In addition to those three projections, I will be including the difference between the combined projection and the average AV/S for a player at that particular draft position. This gives an idea at what kind of surplus value can be gained by selecting a player where they are projected to go.
As for the quarterbacks, we will be projecting average AV per season for each wide reciever and tight end. Unlike the QBs, we will be looking at all players drafted, rather than just the first three rounds. This is because there are lower barriers to entry for other positions as compared to QB, and we are better able to make projections on lowly drafted players as a result.
Wide Receivers
N’Keal Harry, Arizona State
6.55 AV/S random, 6.19 AV/S linear, 6.37 AV/S combined, 1.32 surplus
Coming in as the number 1 ranked wide receiver is the standout from Arizona State, N’Keal Harry. Harry was a ball hog at ASU, gathering the highest number of catches before age 21 in the whole draft class, in addition to a whopping 47% of his team’s receiving touchdowns last year. Harry also provided some value outside of receiving in his career, getting 144 rushing yards and 165 punt return yards. N’Keal provides a significant amount of surplus value relative to his projected draft slot, and should be a top target for a team searching for a WR at the end of the first round.
2. DK Metcalf, Ole Miss
4.70 AV/S random, 6.21 AV/S linear, 5.46 AV/S combined, -1.51 surplus
DK Metcalf is a good example of why looking at the surplus value number is important. While DK comes in as the 2nd ranked prospect by the models, he is actually significantly lower than what you would expect from a player currently projected to go 10th overall. While DK is a size/speed freak, his below average production should give teams serious pause. While I won’t take Julio Jones-like potential off the table, I would rather trade down and take N’Keal Harry (or a couple of other guys upcoming) than roll the dice with Metcalf in the top portion of the draft.
3. Andy Isabella, UMass
5.49 AV/S random, 4.76 AV/S linear, 5.13 AV/S combined, 2.17 surplus
The models are obsessed with Andy Isabella. Projected as a late day 2 pick, Isabella is an undersized track star from small school UMass. Isabella was the most productive prospect in this wide receiver class, gathering 4.13 yards per team attempt and 0.031 TDs per team attempt. While he is slightly older than maybe you would want, Isabella had the ability to completely take over a game at the college level. His transition to the NFL relies on finding a true role in an offense and overcoming his small stature, but the models think he will find a way to succeed in spite of that. Isabella is a guy I would not want to leave the draft without selecting if I was a team that desperately needs a WR like the Ravens, Redskins, or Patriots.
4. AJ Brown, Ole Miss
4.10 AV/S random, 5.31 AV/S linear, 4.71 AV/S combined, -0.34 surplus
Long time producer at Ole Miss, AJ Brown has been a projected high draft pick for a long time. While people have become more hyped about his teammate DK Metcalf, Brown still projects to be a fine receiver at the NFL level. His production has been good to great in all aspects, with the sole exception of touchdowns. While he does have a negative projected surplus, the difference is fairly small and would not prevent me from taking him at his projected draft slot if my team likes his fit with our team.
5. Hakeem Butler, Iowa State
3.55 AV/S random, 4.84 AV/S linear, 4.20 AV/S combined, -0.44 surplus
Full disclosure: I LOVE Hakeem Butler. When I watch his tape, I get total AJ Green vibes and see him as a dominant WR1 in the NFL. However, the model has more reservations than I do. Hakeem was a late bloomer in college, which historically have had a higher rate of NFL busts than guys that produce early on college. However, his last season was sensational, with competitive metrics with any other prospect in the class. In the end, Butler looks like a boom or bust prospect to me, which I tend to think with boom. Like Brown, he has a slight negative surplus but it’s a roll of the dice I might be willing to take at the end of the first round.
Sleeper: Ashton Dulin, Malone
4.11 AV/S random, 3.41 AV/S linear, 3.76 AV/S combined, 3.28 surplus
Ashton Dulin may be the best kept secret in the draft. After a four year playing career at Malone University, Dulin came onto my radar after smashing the combine where he ran a 4.43 40 time at 215 lbs with great jumps. Digging a little deeper, I found that Dulin had unreal production at the college level at all levels: the receiving game, the running game, and on special teams. While he played at an insanely small school (Malone actually folded their football team after this year), Dulin has metrics that would suggest he can be a legitimate NFL starter. Boasting the highest surplus value at a projected selection at the very last pick of the draft, getting Dulin should be of the highest priority for any club this year.
Tight Ends
TJ Hockenson, Iowa
3.66 AV/S random, 5.69 AV/S linear, 4.68 AV/S combined, -0.92 surplus
The consensus top tight end in the class also lands as the model’s number 1 TE, but a much weaker one than typically ranked. While he had great production and comes into the NFL as a 21 year old, Hockenson’s weight adjusted speed is only slightly above average. Personally, despite the very low surplus of -0.92, I think Hockenson will end up as a pretty good tight end, but his ceiling may not be as high as the other first rounder in this class….
2. Noah Fant, Iowa
4.10 AV/S random, 5.08 AV/S linear, 4.59 AV/S combined, -0.01 surplus
Noah Fant is a size/speed unicorn. Out of players drafted in the first three rounds used in the model, only Evan Engram has a higher speed score. Combined with solid production, Fant is a legitamte top tier tight end prospect.
3. Caleb Wilson, UCLA
3.46 AV/S random, 3.45 AV/S linear, 3.46 AV/S combined, 0.83 surplus
Uber productive in college, UCLA’s Caleb Wilson is the premier sleeper in this year’s tight end class. Wilson garnered a class leading 35.7% of his team’s receiving yardage, a number that rivals many wide receivers. If I’m a team needing a pass game threat at tight end, I would pencil in Caleb Wilson as my selection in the third round.
4. Irv Smith, Alabama
2.81 AV/S random, 3.79 AV/S linear, 3.30 AV/S combined, -0.40 surplus
Irv Smith is a solid, if unspectacular tight end option. With average metrics across the board, Smith grades out fairly well, but nothing that jumps off the page. I would probably pass on him at his second round price tag, opting to take an elite option in Hockenson/Fant or waiting a round and taking Wilson.
5. Josh Oliver, San Jose State
1.84 AV/S random, 2.77 AV/S linear, 2.31 AV/S combined, 0.37 surplus
After Smith, there is a fairly steep decline in expected talent in this tight end class. While Oliver provides decent value for his draft slot, he isn’t expected to be much more than a backup level tight end. While he isn’t a terrible selection in the 4th, I would rather pay up a bit and take one of Hockenson, Fant, or Wilson at a higher price tag.
Introduction
The regular season is long over, the super bowl is done with, and it is officially draft season. This year, I want to commit to putting out a comprehensive analytics guide for each major position group. Using multiple predictive models, I want to go in depth with the top 10ish players for every position. Today, let’s start with the quarterback prospects:
Quarterbacks, at least for me, have to be evaluated much differently than other positions. And in my opinion, it’s not even necessarily because they are any harder to evaluate, it’s because opportunities for quarterbacks are controlled mostly by narrative and not by ability. Highly drafted QBs are given way more chances than deserved (looking at you Gabbert) and lowly drafted QBs are rarely given a second look. For that reason, much like Football Outsiders’ QBASE, I will only be looking at QB’s drafted in the first 100 picks of the draft. Doing this gives us a better sample of quarterbacks that actually get the chance to play. For this set, we are using QBs from 1990 to 2013.
Model Details
For each prospect, I have run 4 different models. One is a multiple linear regression that projects a prospect’s career per season approximate value. Approximate value is a metric created by pro-football-reference.com that boils a player’s seasonal performance down to a single number. It’s not a perfect metric, but I like using it to compare across different positions. The next three models are three logistic regressions that classify each prospect’s chance at getting 1 or more, 3 or more, and 5 or more seasons of above average era-adjusted adjusted yards per attempt. For each of those, I have set two thresholds to separate the prospects with high risk and the prospects with high upside. The following chart shows the success rates at these thresholds:
|
Model
|
High Upside
|
High Risk
|
1st Rounders
|
2nd-3rd Rounders
|
|
1+ Season
|
~82% Success
|
~21% Success
|
54% Success
|
27% Success
|
|
3+ Seasons
|
~69% Success
|
~15% Success
|
37% Success
|
16% Success
|
|
5+ Seasons
|
~36% Success
|
~10% Success
|
27% Success
|
7% Success
|
As you can see, the high upside and high risk thresholds outperform draft position for the majority of classes. However, the results are even better when you bake in draft position:
Model
|
1st + Upside
|
1st + None
|
1st + Risk
|
2nd/3rd + Upside
|
2nd/3rd + None
|
2nd/3rd + Risk
|
1+ Season
|
~84% Success
|
~43% Success
|
~29% Success
|
~50% Success
|
~35% Success
|
~16% Success
|
3+ Seasons
|
~71% Success
|
~33% Success
|
~20% Success
|
~50% Success
|
~30% Success
|
~12% Success
|
5+ Seasons
|
~45% Success
|
~50% Success
|
~15% Success
|
~20% Success
|
~0% Success
|
~6% Success
|
The big takeaway I found is that the 1+ and 3+ models follow the order of 1st + Upside, 2nd/3rd + Upside, 1st + None, 2nd/3rd + None, 1st + Risk, and 2nd/3rd + Risk. This is how I would expect the success rates to break out, suggesting to me that using the models will help us increase our chances of finding successful QB prospects. Using these we can break down the group of possible QBs down to different tiers, helping set expectations for each prospect. The 5+ model does not follow this trend, and I would chalk that up to a low sample size. Still, we can see the general trend that prospects that fall under the risk group are much more likely to not reach 5+ above average seasons.
Prospect Grading System
Before I take a closer look at each prospect, here’s an example of how I would grade a prospect based on these models. This would be the grade I would assign Baker Mayfield from his data last year: 10.66 UUN. The first number is the result of the linear regression I mentioned earlier, based on Baker’s data we can expect him to average 10.66 AV per season over his career, an extremely high number. The next three numbers stand for the category he falls into based in the categorical models: Upside for 1+ Season, Upside for 3+ Seasons, and None for 5+ Seasons. Overall, Baker would have been a blue chip QB in our grading system. The only piece he could have been better was the 5+ model, and that’s the one we take the least stock into.
2019 QB Grades
Without further ado, let's get onto the grades for this year’s class. For each guy I’ll give his grade, and explain what the model likes and doesn’t like for each guy:
1. Kyler Murray: 8.10 UUU
The highly polarizing Kyler Murray lands as our top graded quarterback, and looks to be an excellent prospect at that. A rock solid 8.10 AV per season and passing all three categorical models means that Kyler is in this draft, or most other drafts. And get this: the 1+ and 3+ season models take into account his paltry size. An early starter at Texas A&M, the models love Kyler’s era adjusted completion percentage, production on the ground, and strength of competition.
2. Dwayne Haskins: 6.47 NNN
Haskins represents a fairly large tier break compared to Murray. A below average score on the AV model, combined with the middling results in the logistic models means that there is some decent risk selecting Haskins early. Haskins had excellent production, but was unable to get a starting job until his Junior year, something all of the models ding him for. While there are exceptions to this, it is very rare for a prospect that breaks out that late to stick in the NFL. A team picking Haskins will hope that he will buck that trend, and that his high level production this year was for real.
3. Drew Lock: 5.67 NNN
I would place another tier break in between Haskins and Lock. Projected to go in about the same area, Lock is significantly lower than Haskins in the AV model. The model loves the fact that Lock was able to secure the starting job at Mizzou in his freshman year, but his production was only slightly above average in his final season.
4. Daniel Jones: 5.45 NNN
Another disappointment, Jones is a similar level prospect to Lock. A long time starter with very bad production, Jones is buoyed by his underrated rushing work, age, and size. While I would not want to select either guy, I would probably take Jones over Lock, as he is projected to go 10 spots later than Lock and has a similar grade.
5. Will Grier: 4.36 NRR
Grier looks like he will be a perennial backup in the NFL. The AV model likes him better than the categorical models, but the capped upside makes him a tough sell. I would pass based on his current projected draft slot of 40. He’s more palatable if he were to slip until the high 3rd round.
6. Jarrett Stidham: 2.68 NRU
Stidham is a lottery ticket. While he is probably not going to turn into anything, Stidham passes the high upside check for the 5+ seasons model. While that model is somewhat unreliable, there is considerable potential upside still. If I’m a team with an already established starter, I would not be opposed to rolling the dice in the late 3rd, getting him working with our OC and QB coach, and hoping he can put the tools together.
7. Brett Rypien: 2.89 NRR
Rypien looks like another backup only type prospect. Experienced and accurate, Rypien’s weak college strength of schedule hurts him. His projected draft slot it 100, which puts him right on the bubble of being considered by the model.
8. Ryan Finley: 1.81 RRR
Finley is, simply put, not a prospect that I would consider. An older, late starter at NC State, Finley put together a respectable season last year, but was not enough to overcome the other negative aspects of his profile, including his very slight frame.
*Disclaimer: Betting advice is just for fun, I am not responsible for any bets placed*
*All bet amounts are based on percent of bankroll, per the Kelly criterion*
Khabib Nurmagomedov (51.18%) vs Conor McGregor (48.82%)
Plays:
2.26% on McGregor at +165
Tony Ferguson (86.98%) vs Anthony Pettis (13.02%)
Plays:
None
Ovince Saint Preux (17.30%) vs Dominick Reyes (82.70%)
Plays:
7.86% on Reyes at -230
Derrick Lewis (25.86%) vs Alexander Volkov (74.14%)
Plays:
5.09% on Volkov at -160
Michelle Waterson (54.96%) vs Felice Herrig (45.04%)
Plays:
1.43% on Waterson at +105
Sergio Pettis (78.00%) vs Jussier Formiga (22.00%)
Plays:
8.55% on Pettis at -150
Scott Holtzman (51.64%) vs Alan Patrick (48.36%)
Plays:
4.81% on Holtzman at +240
Gray Maynard (25.06%) vs Nik Lentz (74.94%)
Plays:
None
Ryan LaFlare (57.76%) vs Tony Martin (42.24%)
Plays:
None
*Disclaimer: Betting advice is just for fun, I am not responsible for any bets placed*
*All bet amounts are based on percent of bankroll, per the Kelly criterion*
Thiago Santos (71.62%) vs Eryk Anders (28.38%)
Plays:
6.53% on Santos at -165
Sam Alvey (91.32%) vs Antonio Rogerio Nogueira (8.68%)
Plays:
32.06% on Alvey at -340
Charles Oliveira (77.86%) vs Christos Giagos (22.14%)
Plays:
None
Fransisco Trinaldo (53.14%) vs Evan Dunham (46.86%)
Plays:
3.56% on Dunham at +168
Sergio Moraes (61.54%) vs Ben Saunders (38.46%)
Plays:
3.05% on Saunders at +245
Thales Leites (50.62%) vs Hector Lombard (49.38%)
Plays:
1.29% on Lombard at +117
*Disclaimer: Betting advice is just for fun, I am not responsible for any bets placed*
*All bet amounts are based on percent of bankroll, per the Kelly criterion*
Mark Hunt (68.46%) vs Aleksei Oleinik (31.54%)
Plays:
9.37% on Hunt at -125
Jan Blachowicz (21.42%) vs Nikita Krylov (78.58%)
Plays:
19.87% on Krylov at -150
Andrei Arlovski (47.78%) vs Shamil Abdurakhimov (52.22%)
Plays:
1.46% on Arlovski at +125
Rustam Khabilov (61.62%) vs Kajan Johnson (38.38%)
Plays:
8.96% on Johnson at +600
Mairbek Taisumov (72.04%) vs Desmond Green (27.96%)
Plays:
2.78% on Green at +420
*Disclaimer: Betting advice is just for fun, I am not responsible for any bets placed*
*All bet amounts are based on percent of bankroll, per the Kelly criterion*
Tyron Woodley (50.38%) vs Darren Till (49.62%)
Plays:
4.01% on Woodley at +130
Zabit Magomedsharipov (68.28%) vs Brandon Davis (31.72%)
Plays:
8.94% on Davis at +850
Jessica Andrade (82.38%) vs Karolina Kowalkiewicz (17.62%)
Plays:
None
Abdul Razak Alhassan (64.90%) vs Niko Price (35.10%)
Plays:
8.50% on Alhassan at -120
Aljamain Sterling (47.74%) vs Cody Stamann (52.26%)
Plays:
6.39% on Stamann at +140
Jimmie Rivera (57.56%) vs John Dodson (42.44%)
Plays:
None
Jim Miller (34.60%) vs Alex White (65.40%)
Plays:
4.49% on White at -150
Irene Aldana (39.32%) vs Lucie Pudilova (60.68%)
Plays:
7.82% on Pudilova at +100
