By Elliot Chin
Ask any fan, player, or pundit for a rundown of the top NFL running backs and Tennessee Titan Derrick Henry will inevitably be at the top of the list. He led the league in rushing yards in both 2019 (1540 yards) and 2020 (2027 yards), and is one of only eight players to have eclipsed 2000 rushing yards in a single season. And despite his potentially season-ending injury, he currently leads the league in yards through nine weeks of the season with 937 yards – 116 more than second best Jonathan Taylor. He handled 82% of his team’s carries, the second highest rate in the league—James Robinson, on the injury-ridden Jaguars, was first in that category. But did the Titans offense capitalize on Henry’s full potential? Was his success a result of over-utilization?
Here, we seek to create a valuation of running backs’ performance on different types of run plays. If we know how effective running backs are when utilized in specific ways, we can assess whether they are utilized correctly.
However, this is a difficult problem to solve. Raw stats alone don’t tell the whole story, and metrics such as Expected Points Added (EPA) can be influenced by offensive line performance, strength of schedule, and game script. EPA is a metric which represents the value of a single play in a drive, measuring the Expected Points after the play minus the Expected Points before it has occurred. Expected Points is found based on a multinomial logistic regression of variables including current field position and time left on the clock, which predicts the probability of each type of score (Touchdown, Field Goal, Safety). The prediction tends to be very accurate. For example, an unexpected 90 yard touchdown run would yield many expected points added, while a one yard nose dive would add almost none. However, EPA does not control for the contributions of each player, but aggregates a whole teams’ performance. To take a deep dive into running back efficiency, and identify if and where Derrick Henry actually underperforms, we turn to Individual Points Added (IPA). IPA is a modification to EPA which attributes points to individual players, allowing greater individualized statistical focus.
IPA is calculated using what’s called a multi-level model. First, we represent each player’s IPA as an unknown variable, then take the EPA of each play from the 2020 NFL season and express it as a point from a normal distribution centered around the summed IPA of every player on the field for that play. Then, we calculate an IPA for each player that minimizes the error in the model’s estimate of EPA. This approach, pioneered for the NFL by Carnegie Mellon University data scientist Ron Yurko, provides an effective metric for player valuations which controls for a myriad of conflicting variables. However, IPA calculations are constrained by data availability—our data, from nflscrapR, includes only play-by-play statistics for skill-position players. Here, we use an IPA model derived from Yurko’s model. With a deep dive into statistics, we can uncover nuanced, unexpected conclusions about running backs.
Results and Analysis
To start, let’s take a look at the average IPA of every running back who had at least 20 carries during the 2020 regular season and playoffs. Here, average IPA is the average change that the running back is responsible for in expected points scored on a given drive before and after a running back gets the ball. For example, scoring a touchdown on 1st-and-goal from the 1 yard line would yield only a modest IPA, as a touchdown was expected anyways. Additionally, if a running back is running well behind a stellar offensive line, their IPA would be average if most other running backs could perform just as well in the same situation. Keep in mind that this analysis of running back carries ignores plays in which running backs catch the ball or block—these plays are also important but outside the scope of our model.
The distribution of running back average IPAs is relatively normal, with a mean of -.0003 and a standard deviation of .0096 among running backs with at least 20 carries. As many in the analytics community have been de-emphasizing the importance of running backs, IPA confirms this trend: being a better quarterback or wide receiver makes a much larger difference than being a good running back. The top 40 quarterbacks have average IPA of .027, with standard deviation .096, while wide receivers with at least 20 receptions have an average IPA of -.004 with standard deviation .070. These discrepancies in IPA are large: a wide receiver or quarterback one standard deviation above his peers has an IPA almost ten times higher than a running back one standard deviation above his peers. At the extreme end of the spectrum, Texans QB Deshaun Watson averaged .26 IPA per throw in 2020, the highest of any NFL QB, while Bills WR Stefon Diggs, 2020’s best WR in terms of IPA, could expect .16 IPA every time he was targeted last season.
The relative lack of importance of running backs is often attributed to the fact that a running back’s performance is heavily dependent on his teammates and opponents. This is why IPA is important: it helps us distinguish between good running backs and running backs with good teammates.
For 2020 running backs, Nick Chubb of the Cleveland Browns led the way, while Miles Gaskin of the Miami Dolphins, a team infamous for their barebones running back corps, unsurprisingly took up the rear. These initial macro-results make sense, but looking at a deeper level and between the lines of the linemen, deeper insights will emerge.
To narrow our scope, we look at how these running backs are used. Every run play corresponds to a run gap, which is traditionally labelled with a letter. A run gap is a location on the line of scrimmage through which a running back can run. Some running backs are better at running all the way along the edge, away from the action but close to the sideline (the E Gap), whereas others excel at running straight down the middle (the A Gap). We classify these run gaps into seven distinct categories.
Normally, run stats aren’t sorted by where a run play happens. However, nflscrapR reports ESPN’s classification of every run from the 2020 season, allowing us to view how each run gap is utilized.
When looking at the NFL as a whole, it is clear that most gaps are used equally, with the exception of the middle gap—which counts for two A gaps.
While there might not be a difference in the usage of run gaps on a league-wide scale, there certainly is a difference between individual running backs. One would expect a big power back like Latavius Murray to be utilized primarily in the A and B (inside) gaps, while a fast and shifty back like Raheem Mostert would be better suited for runs to the C and D (outside) gaps. On a team level, we looked at which teams most effectively distribute carries between their running backs on aggregate.
To answer this question, we compared the average IPA for each run gap between each running back on a given team, looking specifically at two teams whose running scheme couldn’t be farther apart—the Los Angeles Rams and the Titans.
Two Rushing Approaches: Rams vs. Titans
The 10-6 2020 Los Angeles Rams housed three incumbent running backs: Cam Akers, Darrell Henderson, and Malcom Brown. After former 2017 AP Offensive Player of the Year Todd Gurley endured a series of setbacks, leading to his release from the team after 2019, Coach Sean McVay employed a running-back-by-committee approach—where multiple running backs have a significant amount of rushing opportunities—to balance the team’s carries.
While on a single-game basis the back with the “hot-hand” often took the majority of carries, over the course of the season, the three Rams backs held relatively even shares in the run game. Towards the end of the season, rookie Cam Akers was on fire, and enjoyed the bulk of the carries, paving his way to be the future backfield bell cow (although he has missed the start of 2021 after tearing his achilles in July). Overall, Cam Akers had the most carries during the 2020 season of any Rams running back in six of the seven run gaps.
However, when we observe the distribution of carries together with the average IPA per runner, we find that this confidence is likely misplaced. It is not Akers but Darrell Henderson who outperforms the others in six of the seven gaps, a potential sleeper facing the brunt of under-utilization. In fact, Cam Akers was exceedingly mediocre, and among the weakest rushers in the league through the left tackle and up the middle. In Akers’s absence, the Super-Bowl contending Rams are 7-2, with Henderson as the leading rusher and sixth overall in the league in rushing yards. For the past few seasons, the Tennessee Titans have found themselves in a position very different from the Rams: they have one of the easiest decisions in the league when deciding who to give the ball to. At 6’3’’ and 247 pounds, 2020 AP NFL Offensive Player of the YearDerrick Henry outsizes the average linebacker, proving himself to be a defensive player’s nightmare and a no-brainer RB1.
Indeed, Henry netted at least 27 more carries than the Titans’ secondary back, Jeremy McNichols, at any given run gap in the 2020-2021 season. Without a doubt, Henry’s mammoth share of carries shows he is not just coach Mike Vrabel’s go-to guy, but truly rushing royalty in the Titans’ backfield.
But sadly, average IPA reveals that Henry does not rule every single run gap. Notably, he struggles on runs at left tackle, averaging an underwhelming -0.040 IPA. Ironically, this is the only gap where McNichols has a positive average IPA. Henry’s 64 carries at left tackle (his third most through any gap), when compared to McNichols’s meager eight, demonstrates that despite Henry’s prowess as Tennessee’s undisputed rushing king, there is still room for McNichols to shine. Obviously, if McNichols only runs through left tackle, opposing defenses will quickly learn to anticipate the route every time he is handed the ball, but it is worth exploring shifting the current 8:1 ratio of carries between the two backs at left tackle.
With Henry’s recent injury, we may soon see McNichols, or perhaps backups D’Onta Foreman or Adrian Peterson, step into the spotlight—even though they’re unlikely to match Henry’s pre-injury numbers of 117 yards per game. But given the low IPAs of running backs across the board, it is unclear just how much of the Titan’s success can be attributed to him. One thing, however, is evident: Henry, when healthy, is one of the best backs in the league—just not in every run gap. Henry’s struggles at left tackle are not an outlier. Few lead rushers for NFL teams are best of their team in all seven gaps—Nick Chubb, the highest-ranked rusher in the 2020 season, underperformed compared to teammate Kareem Hunt on runs to the left tackle, right tackle, and end.
The Rest of The League
Alvin Kamara was a bit more dominant compared to his former teammate, Latavius Murray, and excelled in all but two run gaps relative to his team. However, his rushing IPA slightly underestimates his usefulness, as his ability to catch the ball on short flare passes is vital to his effectiveness as NOLA’s lead back. Among the Patriots, Ravens, Chiefs, and Buccaneers, there is no clear lead rusher in terms of efficacy, as reflected by convoluted IPA graphs. Similar to how the Rams did, these teams often use a running back committee. But—if JK Dobbins or Gus Edwards were not on the same team, they would stand out. Their weighted average IPA across all run gaps is good for second and third in the league. Despite their injuries this year, expect to see them on the rise in the coming seasons.
IPA, however, does come with its caveats. For instance, although many running backs excel at one or two specific run gaps, it would be unwise for teams to use them solely for those runs — after all, if opposing defenses saw McNichols in the backfield, it would be simple to shift the defensive formation for the inevitable rush to the left tackle. Plus, factors like the strength of a team’s offensive line can also skew data to make running backs appear stronger or worse than they may be with another team; Le’Veon Bell has never reached the same heights after leaving the Steelers 3 years ago. IPA accounts for these differences by calculating weights for different teams, defenses, and play types; however, with limited data (only 5-6 running backs rush behind the Steelers’ line in a given year), the weights are not perfect. But, even with these shortcomings, IPA can prove effective for helping teams craft rushing or passing strategies that effectively utilize their personnel. The ability of IPA to screen out external factors better than most other football statistics means it provides a more accurate intrinsic view of running backs. This assessment of running back value, when combined with evaluations of offensive lines and defenses, can yield predictive power in analyzing rushing plays.
Overall, statistics provide a deeper understanding of the nuances of football that fans and even coaches often miss. IPA reveals new insights into Derrick Henry’s strengths and weaknesses. Perhaps even more importantly, however, IPA can help reveal “diamonds in the rough”—runners like JK Dobbins and Gus Edwards, who aren’t fan favorites but incredibly effective nonetheless.
Elliot Chin (’25) is a freshman at Harvard College. If you have any questions about his article you can email him at firstname.lastname@example.org. This article was written along with Alex Chen (UCLA), Noah Van Horne (Georgetown), Quetz Medina (Brown), and Joseph Kraus (Stanford).