One of the most shocking trades in NFL history happened right before this year’s regular season. After receiving a sudden trade request from star edge rusher Micah Parsons, the Dallas Cowboys traded him to the Green Bay Packers for two first round picks and veteran defensive tackle Kenny Clark. Simply put, this move stunned the entire pro football world; Parsons has long been one of the best defensive players in the league, and perhaps even the face of the Cowboys.
There’s still a long way to go until the Super Bowl, but this move seems to have benefited all parties. Initially entering the year with a weakness in pass rush, Green Bay has benefited from the presence of Parsons, who is currently second in the league in total pressures. While the Cowboys defense has mostly struggled, the team stands at a competitive 6-6-1. Weirdly enough though, this wasn’t the only high-impact move Dallas made this season, as right before the trade deadline, they ended up indirectly replacing Parsons with ex-Jet interior stud Quinnen Williams in yet another blockbuster move. Though Dallas’ playoffs odds are barely hanging by a thread, they still have draft capital for the future and remain a competitive team.
Whether it’s in the Cowboys decision to trade Parsons or to leverage the assets gained from the Parsons-trade to swing for QIlliams, I’ve always wondered about the value of elite defensive players. What significant differences in value exist between strong interior defenders and edge rushers? Is any such difference reflected by the market? If so, is the market right or wrong?
In Part 6 of my Untangling the NFL series, I built a predicted EPA model for offensive linemen using Pro Football Focus grades as my main independent variable. For today’s focus, I’m shifting my focus to defensive players. This time, I was able to avoid PFF grades entirely and rely on their underlying statistics instead. With that said though, what I ended up with was quite surprising.
The Unique Defensive Attribution Problem
While EPA isn’t a perfect measure of individual impact, it’s largely useful for capturing a player’s productive value in their specific area of contribution. Yes, you might quibble with Brock Purdy looking like a top-five quarterback, knowing that much of his statistical production comes downstream of his supporting cast and system. But at the very least, there’s a clear logic to EPA: the ball goes to a player, something happens, and we can attribute the result to someone, even if their success is shaped by what people away from the ball happen to accomplish.
Defense suffers from a similar issue, but on steroids and with a lot more noise. Unless you’ve forced a turnover and hold a ball – one of the rarest occurrences in a typical play – it’s very difficult to attribute points to defensive outcomes. Furthermore, most research in pro football analytics seems to back the fact that sustaining defense over a long period of time is more difficult than offense in large part. Even individual statistics like tackles, sacks, and interceptions are tough to apply context toward when they’re uniquely more entangled than their offensive counterparts.
For example, how can you quantify the difference between a “coverage sack” vs a “pressure sack?” Are tackles really a good indicator for strong defensive performance, or are they a sign that someone is getting beat a lot? For now, answering these types of questions on a case-by-case basis is beyond the scope of my project. But it doesn’t mean I can’t try a different, more broad, but still useful approach.
An Empirical Approach
Much like I did with offensive linemen, I turned to PFF, which meticulously tracks snaps, pressures, sacks, tackles for loss, run stops, and coverage metrics like yards allowed, interceptions, and pass breakups. Using these as season-level aggregates for individual players, I cross-referenced this data with nflfastR play-by-play EPA values to derive empirical weights.
First, I mapped each player to one of five defensive roles based on snap counts and pre-snap alignment: EDGE (edge rushers), IDL (interior defensive line), LB (linebackers), CB (cornerbacks), and S (safeties). From there, I developed a two-bucket allocation structure based on EPA magnitude analysis of my entire play-by-play data. I ended up weighing pass defense at 69 percent and run defense at 31 percent. For each of these categories, I ran statistical analyses on different statistics’ percentage “share” of EPA.
Pass Defense
| Metric | Weight |
| Coverage Snaps | 35% |
| Yards / Coverage Snap (efficiency) | 25% |
| Pass Breakups | 15% |
| Sacks | 11% |
| QB Hits & Hurries | 4% |
| Interceptions | 10% |
Run Defense
| Metric | Weight |
| Run Stops | 64% |
| Tackles for Loss | 27% |
| Forced Fumbles | 9% |
A key addition in this model is the yards-per-coverage-snap efficiency metric. Rather than just rewarding volume (more snaps = more credit), I inverted the yards allowed metric so that players who allow fewer yards per snap receive proportionally more credit. This helps distinguish between a cornerback who plays 800 snaps and allows 600 yards versus one who plays 800 snaps and allows 1,200 yards.
To weight run defense contributions by position, I analyzed where each position typically makes their tackles. Edge defenders average -0.6 yards (behind the line), while linebackers average 1.7 yards downfield. This informed position multipliers applied to stops and TFLs: edge defenders receive 1.07x credit relative to interior linemen, while linebackers receive only 0.21x—reflecting that a linebacker’s “stop” often occurs after more damage is done.
I should note here that you may have noticed that tackles are currently unaccounted for in our model. That is due to scope limitations and time; for the purpose of this project, I’m treating any tackles beyond 4 yards as a “negative” play and something to not reward. Obviously, as a football fan, I understand that the game is not this simple. With more time, I would like to be able to discover a way to properly evaluate “mitigation” for defenses rather than just positive plays. But for now, I’m only examining stops, tackles for losses, and forced fumbles in this area of the game (due to the difficulty of filtering fumbles by pass plays and rush defense plays).
For coverage, linebacker coverage snaps were discounted to 0.81x of cornerback value due to the average catch rate disparity between linebackers (79.5%) and cornerbacks (64.6%). Meanwhile, safeties received a 0.94x multiplier reflecting their intermediate catch rate.
Addressing Team & Positional Context
Sometimes, a rising tide can lift too many boats. A key issue that happens with allocation-based models is that average players on elite units can have inflated values, while highly productive stars in bad situations can be needlessly punished. To overcome some of these concerns, I used a multi-layered regression approach, regressing each team’s WAR pool 50 percent toward league average before allocation.
After this step, I applied position-specific regression factors at the individual level. Truthfully, this was a bit of book-cooking, but I thought it was justified because of the difficulty that comes with evaluating positional value. For cornerbacks and safeties as examples – places where scheme, supporting help and quarterback pressure play heavy roles in shaping defensive outcomes – it’s much harder to isolate individual contributions.
For what it’s worth, PFF’s research in WAR seems to provide some semblance of support for my approach, as their team’s year-to-year correlation data showcases marginal levels of repeatability for cornerbacks and safeties. Conversely, edge rushers are much more reliable. In line with PFF’s numbers, I regressed the values of positions further away from the ball (like safeties and cornerbacks) toward their positional means.
The last step was to adjust for cornerback roles on the same team. This was tricky because in some cases, slot corners or ‘number two’ corners had more ‘efficient’ statistics than top corners who may have received more snaps but had ‘worse’ statistics because of their higher caliber of assignment. To adjust for this, I used snap share as a proxy for determining coverage hierarchies in a secondary, applying slight boosts to cornerbacks with the most number of snaps (my model’s ‘guessing’ method for number one corners) and slight decreases for other corners.
Positional Results
At the end of my research, I had 3,293 seasons worth of data. In line with my minimum snap criteria for offensive positions, I set a minimum filter of 50 defensive snaps from 2021-2024.
| Position | n | Avg WAR | Std Dev | Max WAR |
| EDGE | 697 | 0.18 | 0.26 | 1.50 |
| S | 547 | 0.15 | 0.21 | 0.96 |
| CB | 744 | 0.13 | 0.19 | 1.10 |
| LB | 551 | 0.12 | 0.19 | 0.99 |
| IDL | 754 | 0.10 | 0.17 | 1.14 |
My research showcases that edge rushers generate the highest average WAR (0.18), followed by safeties (0.15), cornerbacks (0.13), linebackers (0.12), and interior defensive linemen (0.10). I found these results to be pretty fascinating in comparison to what I discovered for offense.
In similar fashion, the further away someone is from the ball at the point of snap, the more value they have as defenders – however, for edge rushers, who apply pressure to the quarterback, they are the exception to the rule because they’re the ones with the closest proximal oppositional impact to the most important position in the game.
At the moment, I think this is a functional model for typical value on an NFL defense. However, I can’t help but wonder if the ‘heavy regression toward mean’ approach used for cornerbacks and safeties currently ‘overcorrect’ some of the positional attribution shares for value. Without them, my findings would have suggested that cornerbacks and safeties were worth roughly one-and-a-half times of value than what my current model proposes. As a result, I’m not very confident in the specifics of these findings.
Four-Year Cumulative WAR Leaders (2021-2024)
The top of this list is dominated by elite edge rushers—exactly what you’d expect if sacks are the most individually attributable defensive statistic. Myles Garrett leads at 2.69 WAR, followed by Cameron Jordan (2.60), Nick Bosa (2.57), and Micah Parsons (2.56). These are consensus top-tier pass rushers, which provides strong face validity for the model.
With that said though, I was caught off guard by the top name at cornerback: Paulson Adebo (1.91). Simply put, if you asked 32 NFL executives whom the top corner of the 2020s was, Paulson Adebo might be the 15th choice. In reality though, I think his ranking reflects a limitation of the model’s emphasis on cumulative contributions and consistency per snap over pure volume statistics. I will have to do more work in the future to figure out how to attribute secondary success to individual players.
Validation Against PFF and Market
To validate my methodology, I compared my positional value estimates to both PFF’s 2018 WAR study and actual NFL market salaries.
| Position | My WAR | PFF WAR | Mkt Implied WAR |
| EDGE | 0.18 | 0.06 | 0.18 |
| S | 0.15 | 0.23 | 0.11 |
| CB | 0.13 | 0.23 | 0.12 |
| LB | 0.12 | 0.11 | 0.10 |
| IDL | 0.10 | 0.06 | 0.14 |
My edge rusher values nearly perfectly match market-implied WAR (0.18 vs 0.18). This suggests the market’s premium on pass rushers is well-calibrated to their actual production.
With that said though, at least intuitively speaking, I have to admit a bit of a pause when it comes to PFF’s implied evaluation of coverage players. Although their team gets a boost from the presence of using PFF grades as a grounding tool (which I have forbidden myself from using outside of last week’s column), I still wonder if they are closer to properly pinning a number to cornerback and safety play than I am; as it stands, this is one of many big limitations of my research.
Conclusions
Without getting into the specifics of his contract, Parsons is currently set for $186 million over four years. According to my model, he’s accumulated 2.56 WAR over the prior four seasons, which puts him fourth among all defenders. The market numbers on Parsons right now – as well as players of his caliber – suggest that in order to technically justify his current contract, he would need to produce roughly 4.1 amount of WAR in this time period. In other words, by pure statistics, they’d need Parsons to be in the running for best defensive player in the league every single season here on out. Although Parsons is relatively young, any reasonable person could assume that this is a bit of an overpay with respect to what the team can actually expect from him.
Obviously though, this doesn’t make a contract bad. Even if Parsons does not perform at a rate that justifies the entirety of his current contract, all they need him to do is contribute at a high enough level to help them in the playoffs. Any future cap analysis with respect to WAR would need to take immediacy into account. In other words, if Parsons plays at a good enough level to help Green Bay win the Super Bowl this year, his contract is a justifiable overpay if he plays like the best player at his position for just one of those seasons (roughly in the ball park of the 1.50 WAR range for edge rushers).
In either case, I think I’ve done enough attempting to encapsulate defensive value in the NFL with available data – at least with this week. Although I don’t think I’ve ‘completed’ this project with regards to properly assessing some of its more glaring limitations (evaluation of cornerbacks in particular), I am satisfied with its baseline numbers and, more importantly, am too exhausted to carry out any further analysis. Expect plenty of post-publish updates.