Vikings

2017 NFL DRAFT: Pre-Combine Production Profiles (Edge Defenders—"4-3" projection)

We’ve been looking at ways to improve draft predictions by incorporating college production into our analysis. First, we looked at offensive linemen, which was largely speculative. The early returns on that analysis are good—there’s a strong relationship between team-adjusted scores for offensive linemen last year and PFF grade this year. Also, Willie Beavers was the worst-scoring tackle and he was the highest-drafted player at any position to be cut last year.

We also looked at linebackers, where we have a stronger history in using predictive production data. The tools used there—team market share of tackles and team market share of tackles-for-loss—seemed like they could be useful for predicting the success of pass-rushers heading into the NFL.

Nope.

Part of the reason is likely because tackles (like another statistic—receiving yards—for a position that has strong predictive history, wide receivers) are zero-sum. Most plays end in tackles, so there’s a limited number of tackles to go around. Being able to produce (or consume?) most of those tackles on a team is indicative of a high-quality player.

That’s not true for sacks and pressures from defensive linemen. If one of the four defensive linemen doesn’t get a sack, there’s a very low probability that someone else will end up with the sack—or pressure. Those are self-produced, so a market-share approach isn’t helpful.

So, what if we used total sacks and tackles-for-loss?

It took me some time to realize that market share didn’t matter, especially because draft-year production in sack totals correlated about as well with NFL outcomes as market share. I was ready to give up on this approach for pass rushers.

After all, Vikings fans are aware that their super-young phenom pass-rusher, Danielle Hunter, only produced 1.5 sacks in his final year of college. Honestly, the previous market share approach does a comically poor job in general. If we produced 100+ style ratings for pass rushers like we did for offensive linemen and linebackers, we’d end up with scores of 90.7 for Everson Griffen, 96.2 for Danielle Hunter (he had a humongous age boost) and 65.1 for Brian Robison.

A Bad Decision, According To the Old Model. Photo Credit: Kyle Hansen (Cumulus Media)

If we looked at the last eight All-Pro edge rushers (Khalil Mack, Vic Beasley, Von Miller, J.J. Watt, Mario Williams, Robert Quinn, Justin Houston and Elvis Dumervil), we come upon a similar problem—they have scores of 87.1, 70.1, 92.6, 96.3, 110.1, 111.9, 101.7 and 99.9, respectively.

That’s one score embarrassingly lower than 85, one close to 85 and none above 115. There’s an average score of 96.2. The All-Pros had an average score below the average.

Players like Von Miller, Brian Robison, Bruce Irvin, Carlos Dunlap and notably Myles Garrett didn’t have their most productive seasons in their final year in college. The year prior was their money year, and using “peak” year instead of final year turned this approach from useless to predictive—without accounting for draft position, a combination of age, production (in the form of sack and TFL totals) and athleticism correlates with NFL outcomes extremely well.

When combined with draft position, it provides better predictive power than the production metrics we have for most positions.

The model did a good job predicting the failures of first-round picks Gaines Adams, Lawrence Jackson and Bjoern Werner.

It did have misses, like Jamaal Anderson, Da’Quan Bowers and Tamba Hali, but its hit rate against the draft is over 60 percent. In other words, when the model argues that a player is over- or undervalued relative to their position in the draft, it is right over 60 percent of the time—according to both sacks per game and career approximate value.

That means that the draft, which generally produces hits (players who outperform their draft value) 45 percent of the time at edge rusher, doesn’t do as well at identifying talent as well as combining pre-draft evaluation with the production model does—which produces hits 73 percent of the time.

There’s a good chance those kinds of results come from the hasty work I did, and the true hit rate is closer to 65 percent. The problems include overfitting, small samples, a lack of random trials and a somewhat dubious output measurement (career approximate value per year, explained here—as one odd example, Jadeveon Clowney counts as a “hit” in CAV per year) but the point was to see if production could help predict performance, not how precisely one can do so. To that extent, I think it has demonstrated that.

This model could probably be improved by taking into account rate—essentially, figuring out how many snaps players played against, instead of using season totals, but this is fairly powerful. As it is, I’m adding in Pro Football Focus’ Pass Rusher Productivity (a rate stat) to the results to help balance out those problems.

And hey, maybe there’s something in the analytically-minded Cleveland Browns drafting the Nos. 1 and 3 sack producers in the FBS last year, as well as the No. 6 tackles-for-loss creator.

With that in mind, let’s look at the production profiles of edge defenders.


Let’s look at the edge defenders typically projected to play as a 4-3 defensive end (and Solomon Thomas). We’ll look at “3-4 outside linebackers” another day. As always, the ranks listed below are taken from CBS’ draft prospect rankings; unlike last time, I’ve updated the player rankings to match to CBS’ rankings at time of publish:

There aren’t any defenders here who have a score above 120 or below 80, so from that perspective, there aren’t too many standout cases. The closest to 120 hail from the state of Florida, so turning an eye there may produce value.

The first is DeMarcus Walker from Florida State. Half a year younger than most prospects, he derives most of his value from his fantastic sack production. With 16 sacks, he leads the draftable class in sack production while also putting together an excellent number of tackles-for-loss.

The only thing to be worried about his production metric—and it’s more a caution signal than a red flag—is that his pressure rate is not commensurate with his sack rate. In other words, he produced many more sacks than one would expect from the amount of pressure he created, which strongly implies regression, even if he’s a good finisher.

With about nine edge rushers ahead of him on CBS’ chart, it’s plausible he could be available at the top of round three while showcasing first-round talent.

The other player is Trey Hendrickson from Florida Atlantic. His 13.5 sacks are impressive, and aside from players outside of the top 150, rank second in the class. Hendrickson diverges from Walker in two respects, however. First, small details balance each other out—his age gives him a boost, while his average tackle-for-loss production drags him down.

Instead, the second difference is more intriguing; his sack production understates his pressure production. He’s more likely to regress upwards than anything else, and his ability to push the pocket is statistically the best among the defensive ends and second-best among the edge rushers.

The next-most productive defensive end is Tennessee’s Derek Barnett, whose athleticism is reputed to be poor relative to the class. If that’s the case, it will end up hurting his grade significantly, as athleticism has the biggest influence of the three inputs (age, production and athleticism), though not bigger than the other two combined. Still, his ability to create sacks should count for something and so far that something implies that he’s currently being undervalued.

Those three stand out above the rest, and by contrast, there’s one who falls below the average. Though Carl Lawson earns praise from draftniks for his technique, it didn’t give him much of an ability to produce sacks. The Auburn product is not too far below average in any one input, but he’s below average in all of them; which pushes him down relative to the rest of this class.

There are two other players that are clearly below average, but not so much that it’s a huge demerit on their draft grade: Dawuane Smoot from Illinois and Taco Charlton from Michigan. Smoot was pegged as a preseason draft riser, but he didn’t beat his dismal sack peak of 7.0. That’s not because of run-heavy offenses in the Big Ten, either—a number of the outside linebackers from the Big Ten produced better sack numbers, and both of them also had low pressure rates, so even when accounting for the number of opponent passing attempts Smoot and Charlton didn’t produce much.

Beyond that, both also produced a poor tackle-for-loss total, which would theoretically go up in a run-heavy slate. The only bonuses either receive are from their age, and it’s fairly minor.

Neither end up having a production profile too different from Lawson’s, except that both are younger.

Tarell Basham from Ohio is an interesting case. He plays for a Group of 5 school (MAC) instead of a Power 5 school, which should matter in a way that it didn’t for the linebacker or offensive lineman because we’re not using market share statistics. I’ve found that incorporating FCS scores damages the predictiveness of the model (at least without some sort of adjustment), so there might be a similar minor effect for prospects from schools that don’t have the level of competition that a Power 5 school has.

Basham’s sack total isn’t amazing outside of that context and somewhat worrisome inside of it. Despite that and his below-average tackle-for-loss production, he ends up with a positive production score that only gets dragged down because of his age. The reason for this is his incredible pressure rate which overshadows both the other statistics.

Whether or not he translates may depend on what you think of his performance against Tennessee which I wasn’t too impressed by despite his sack and 1.5 tackles-for-loss.

There’s no production information for Tanoh Kpassagnon from Villanova or Derek Rivers from Youngstown that are useful in comparison, but by themselves can be somewhat revealing. I don’t have pressure rate for Kpassagnon or Rivers, but we do know that the Villanova product produced 21.5 tackles-for-loss and 11 sacks in his final year, which is pretty good but could be better given whatever theoretical adjustment we would want to make for FCS prospects.

That said, having talked to Kpassagnon at the Senior Bowl, I was surprised to learn that he was often asked to play a “4i” technique inside the shoulder of the tackle, instead of on the outside. That likely depressed his production, and might be interesting to explore further.

Rivers had 19.5 tackles-for-loss and 15 sacks but in more games than anyone else in the set, including the national championship contenders in the outside linebacker set. Rivers’ production isn’t anything to write home about and both Rivers and Kpassagnon also have some age issues relative to the rest of the class.

The rest could probably be considered to have functionally average production scores. That doesn’t mean one should downgrade Myles Garrett relative to the rest of the class; instead one should simply stick to the film grade until athleticism scores come out. Given that the consensus on him seems to be “top of the draft” it really doesn’t matter how far out positively he grades, only if he ends up grading clearly negatively (below 95).

It seems unlikely that the combine will hurt him too much, however. Even knowing that it plays a bigger role than age or production individually, the fact that both of those elements are ahead of the pack should buffer him against average or even below-average combine results.

Charles Harris from Missouri is sometimes talked about as if he’s getting lost in all the solid edge prospects, but he also receives talk that his run defense might make him a designated pass-rusher only (coincidentally, the only fit people might have seen from fellow Missouri prospect Michael Sam).  If so, his profile overall doesn’t look good; his sack production and his pressure rate are both below average among the draft prospects (in fact, his pressure creation by itself ranks at an 86).

He only makes up for it in age and, weirdly, his tackle-for-loss production. If run defense is an element of his game that doesn’t translate to the next level, then there are some serious concerns about what kind of player he can be.

In this case, Harris is the kind of player whose average overall score might be more disconcerting rather than neutral.

Then there’s Solomon Thomas, who many project to play inside but enough project to play on the outside that I included him. I’m not sure what to do with him in that capacity, but his production shouldn’t hold people back at this point.


Next time, we’ll tackle those edge rushers projected to be outside linebackers. Even though the Vikings are sticking with a 4-3 system, it remains relevant to do this because players like Danielle Hunter were projected to play 3-4 outside linebacker (and at various times, Brian Robison and Everson Griffen) and produced for the Vikings regardless.

As it stands, the combine will be interesting for players like Charles Harris and Tarell Basham, where a good score moves them out of danger territory. It will have an impact on Derek Barnett that could re-center his grade, and likely won’t have much of an effect for Myles Garrett. The biggest risers to watch are those players from Florida (DeMarcus Walker and Trey Hendrickson) who could jump past 120 and might therefore deserve serious reconsideration.

Vikings
The Vikings Are Getting Caught In the NFL Draft Silly Season Vortex
By Tom Schreier - Mar 29, 2024
Vikings
Finalizing the Roster: New Signings, Contracts, and Draft Day Rumors
By Matt Johnson - Mar 28, 2024
Vikings

Did the Vikings Ultimately Come Out Ahead In Free Agency?

The Minnesota Vikings may have had the most active and impactful free agency session in the NFL, even though they didn’t sign as many new players as […]

Continue Reading