One of the biggest storylines of the 2025 NFL season has been the rise of the Indianapolis Colts. Leading the charge is running back Jonathan Taylor, whose totally dominant run production has sparked a growing MVP narrative. If he won, he’d be the first non-quarterback to take home the award since Adrian Peterson in 2012.
It would be controversial, but zooming out a bit, the NFL seems to be going through a skill position renaissance. Second-year tight end Tyler Warren, Taylor’s very own teammate, has broken out as a true dual-threat as a run blocker and a threat in the passing game. Meanwhile on the West Coast, Jaxon Smith-Njigba has put together a historic first half of the season, already surpassing 1,000 receiving yards. And who can forget last year, when Saquon Barkley had arguably the greatest running back season in NFL history?
Given the nature of my semester-long project (encapsulating value in football analytics), I find these developments especially pertinent. Last week, I built a QB WAR model; this week, the focus shifts to the skill positions: running backs, wide receivers, and tight ends.
Methodology
Like I did with my previous column, I used four NFL seasons worth of play-by-play data, going from 2021 to 2024. Making sure to account for the same plays that I excluded last time (special teams, spikes, kneel-downs, and penalty-nullified snaps), I had hundreds of thousands of plays, this time split into passing and rushing plays Since nflFastR tracks Expected Points Added (EPA) per play (EPA), I made sure to track these statistics and apply them to every skill position player The idea ended up being largely identical to last week’s quarterback model.
There are a few key differences between how I assessed skill position value from quarterback value. In last week’s post, I wrote that I did not use raw EPA for quarterbacks – instead, I used a modified version of EPA that was averaged with “modeled EPA” as predicted by Completion Percentage Over Expected (CPOE) and Success Rate. I did this in order to have more stable results, since quarterbacks constantly touch the ball and are scarce in availability.
That is not the case for skill position players, where they largely have far fewer touches and much narrower scopes of influence. Although I theoretically could have tried to balance raw EPA with something like ‘modeled EPA” adjusted for Rushing Yards Over Expected (RYOE) or Catch Rate Over Expected (CROE), for now, I’ve just estimated value through the lens of raw EPA.
As you read through my results, keep in mind three large caveats. The first one is the fact that the entire EPA of a rushing or receiving play is attributed to the ball-carrier. I did this mostly due to a limitation of public data related to lineup information for plays and blocking statistics. In a later piece, we will deal with this issue; or now, accept it as a limitation of our methodology.
The second caveat, however, is the more liberal minimum plays threshold I have set for skill position players. While I have defined replacement-level as the bottom 25th percentile (and below) of EPA, I have instead set a minimum of 25 touches instead of 100 passing/rushing players for quarterbacks. I picked 25 out of a baseline “guess” of how much less frequently the typical skill position player touches the ball than quarterbacks.
The third caveat is that unlike with quarterbacks – in which we calculated QB EPA through taking EPA and multiplying it by the percent of plays involving the quarterback and the percentage of total offensive EPA attributed to QBs – we had a far simpler process to calculate skill position EPA. This process ended up simply being Offensive EPA per win (the number I detailed in the last column) minus QB EPA per win distributed to all skill position players. This doesn’t take into account lineman or blocking contributions, but in a future piece, I will explain how I have chosen to deal with this part of the game.
Initial Insights
| Cumulative 2021-2024 Table (Sorted by Total WAR) | ||
| Player (Position) | Plays | Total WAR |
| Justin Jefferson (WR) | 431 | 2.89 |
| Mike Evans (WR) | 346 | 2.80 |
| Jamar Chase (WR) | 465 | 2.78 |
| AJ Brown (WR) | 356 | 2.67 |
| Tyreek Hill (WR) | 498 | 2.53 |
| Amon-Ra St. Brown (WR) | 483 | 2.51 |
| CeeDee Lamb (WR) | 497 | 2.42 |
| Davante Adams (WR) | 423 | 2.29 |
| Tee Higgins (WR) | 294 | 2.29 |
| Amari Cooper (WR) | 278 | 2.16 |
To begin with the obvious: wide receivers completely dominate our EPA-centric foundation. This is in part because they’re only measured on plays that involve them, but also because they typically have more yards per play than other skill positions, due to them stretching the field. In fact, it’s only after Terry McLaurin (2.16) – the Washington Commanders wideout – that a non-WR shows up in our Top 25 cumulative WAR ranking: running back Derrick Henry (2.12) at No. 12. After are two fellow non-WRs in George Kittle (2.00) and Travis Kelce (2.00) at No. 13 and 14. Mark Andrews (1.68) is the only other non-WR in the Top 25.
To examine if this split held true across the entire league and not just our Top 25, I ran a box plot comparing the typical distribution of WAR between running backs, wide receivers, and tight ends. I ended up finding out that far from just the outliers, of which many of them are actually close together, the trend holds steady even for the typical player.
In total, we had 400 running backs, 167 tight ends, and 431 wide receivers, but each of the average and median total WARs highlighted notable differences. Each of these for running backs (0.05 and 0.03) and tight ends (0.15 and 0.12) were significantly smaller than they were for wide receivers (0.29 and 0.25). Although some of this can be explained by blocking value being underplayed in my EPA-heavy methodology, I believe it’s a reflection of current trends in the NFL, where spacing and stretching the field have a grweater premium than winning trench fights. My guess is that even if blocking were to be factored in, wide receivers would still hold a huge advantage because of their unique per-touch value.
But while the average receiver is more valuable than the average tight end or running back, it does make the excellent non-WRs stand out that much more. Take 2024 for instance: a year in which each of the top three spots were held by running backs: Derrick Henry (1.10), Saquon Barkley (1.03), and Jahmyr Gibbs (0.84). Considering any minimal gain in WAR that could come from blocking could further contextualize the contributions of dual threat tight ends as well, with players like Kelce and Kittle being relatively underrated by EPA. This would suggest that while wide receivers largely dominate statistical production for offenses, running backs and tight ends at their best can match them and perhaps have even greater importance due to standing above their peers even more.
Although I initially compared skill position player value against each other flatly, I was also curious to see how they compared with quarterbacks. Below is a table of information related to the typical quarterback in the NFL and the typical skill position player – particularly when it comes to average and median WAR for each position.
| Position Group | Sample Size | Minimum Snaps Threshold | Average Total WAR | Median Total WAR |
| QB | 209 | 100 | 1.05 | 0.71 |
| HB (including FB) | 400 | 25 | 0.06 | 0.03 |
| TE | 167 | 25 | 0.15 | 0.12 |
| WR | 431 | 25 | 0.30 | 0.26 |
Again – there are certainly inconsistencies to note here, between the divergence of minimum snaps thresholds, sample size per position, and the fact that these are obviously just raw numbers. But in the next section, similar to what I did with quarterbacks, I’m going to run a slightly different validation check than the ones I performed last week.
What Does PFF WAR Tell Us?
In the late 2010s, writers from Pro Football Focus submitted a research paper to the Sloan sports conference detailing their own methods for calculating WAR. Without diving into every single detail, there are many significant differences between my methodology and their methodology, but the three biggest ones involve their incorporation of PFF grades into their evaluation process, their greater data set, and increased minimum snaps threshold (250 across all positions). With that said though, I wanted to compare my findings with PFF’s findings, as they still remain experts in the field.
| Position Group | Sample Size (Me) | Sample Size (PFF) | Average Total WAR (Me) | Average Total WAR (PFF) | Net Diff |
| QB | 209 | 994 | 1.05 | 1.63 | -0.58 |
| HB (including FB) | 400 | 2373 | 0.06 | 0.10 | -0.04 |
| TE | 167 | 1621 | 0.15 | 0.18 | -0.03 |
| WR | 431 | 2864 | 0.30 | 0.28 | +0.02 |
The only place where I believe the divergences are noteworthy are in quarterback evaluations. Remember that I deliberately took a conservative approach to measuring replacement-level (bottom 25 percentile of quarterbacks with 250 snaps), while PFF measures replacement-level players as the ‘typical starter’ on a three-win team (doing so through advanced math and measuring against projected rather than actual wins). Meanwhile, the divergences between the skill position players are either largely negligible or can be reasonably assumed as a byproduct of blocking (or frankly other positions) not being factored in yet for my version of WAR.
How Does Skill Position WAR compare to Skill Position AV? (2024)
In line with what I did with QB WAR, I’ve compared my methodology for HB, TE, and WR WAR with Approximate Value at each of these positions. To keep our comparison short, I picked only the first page of roughly 150+ players with the highest AV in the 2024 season, using the 2024 seasonal data for WAR as a point of reference.
Among our 167 sampled players, each of the Pearson (0.605) and Spearman correlation (0.486) demonstrate a modest to strong correlation from our WAR model for skill position players to AV. I found this especially notable because WAR, as I’ve set it up, takes into account postseason performance, which AV does not. At the same time though, there remained enough variation between WAR and AV with respect to position.
| Position | Size | Pearson Correlation (WAR vs. AV) | Spearman Correlation (WAR vs. AV) |
| RB | 51 | 0.734 | 0.504 |
| WR | 87 | 0.612 | 0.616 |
| TE | 29 | 0.669 | 0.760 |
| All Positions | 167 | 0.605 | 0.486 |
When it comes to evaluating our sample of players covered by AV and WAR, it’s important to note that this model only explains roughly 36.6 percent of its results. At the same time though, given how I haven’t accounted for blocking yet, I believe that my approach is a bit more precise when it comes to quantifying a player’s true value in terms of their production. To put this to the test though, I ran a final regression check, measuring ‘predicted skill position WAR” of a whole team with actual wins.
At the team-level, I found that we had a relatively strong Pearson correlation (0.582) and Spearman (0.599) correlation, with a model that explains a little over a third of wins variance. Roughly speaking, what this model tells us is that a team of replacement-level skill talent will win only six games in a given season, with each additional ‘team-wide win above replacement’ being worth roughly another two wins.
Out of curiosity, I decided to run our results against AV in similar fashion to what I did with QB WAR – once again, WAR managed to outperform it, though by a hair.
Limitations & Considerations
Quantifying differences in value among skill positions can help teams build better rosters in pro football. The vast importance of wide receivers in our research, as well as our moderate strength signals, seems to indicate that stretching the field and completing deep places are anywhere from twice as important to five times as important as blocking and running the ball.
At the same time though, I think that the biggest hole in the current model comes from its lack of factoring in blocking. Combined with the recent ascendancy of pass-catching running backs,of whom the top three matched or exceeded every other player in the league, there’s reason to wonder if the differences in WAR between the skill positions aren’t going to necessarily hold for the future.
In general, the lack of granular data on blocking does make me worried that my methodology slightly underplays the roles of skill position players as blockers, but heading into next week’s piece on offensive lineman value, I wonder if it’s best to swap from an individual EPA-foundation approach to one that measures “team-wide blocking,” with some portion attributed to entire lines and then allocated to individual linemen based on additional per-game data to complement the per-play data present in nflFastR. I’ll find the answer next week.