In the NFL, from year to year, players often experience a significant fall-off in production, but no player is more affected than the running back. Running backs need to be versatile, with the ability to run and catch the ball, and because they are lining up against defensive players who are the strongest and most aggressive on the field, which makes the position physically demanding. The average running back is just under 6' tall and only 215 pounds while the average defensive lineman is around 6'3'' and nearly 300 pounds which puts running backs at a significant size disadvantage.
There is an emerging view among NFL fans and front office executives that due to injuries and changing offensive-defensive schemes, running backs are being drafted too early, are overpaid, and that the position is becoming increasingly interchangeable. Support for this contention is that the running back is subject to production decreases greater than any other player on the field. In 2010, Chase Kennedy, a student in the economics department of Haverford College, conducted a study regarding the deterioration of NFL running backs, basing his study on the amount of career touches in relation to running back production, quantifying using total touches, yards, and touchdowns. He began his study by finding players who had, at some point in their career, a season with 100 or more touches, and from there narrowed it further to players who had at least six seasons with 100 touches. His final dataset consisted of 96 players, all since the AFL-NFL merger in 1970. Kennedy’s study found that for every 250 carries, a running back experiences a drop off in production.
Using the programming language R, we decided to load the last ten years of NFL play by play data using the nflfastR package, with the intent of conducting a new study that would examine modern-day running back consistency and the "running back wall" that Kennedy described. There were a lot of issues with the data, and extensive data cleaning needed to be done to remove empty data entries, nullify receiver and quarterback rushing data, and more. Furthermore, we filtered the data to only include seasons in which a running back surpassed 100 rushing yards to account for season-long injuries. We created two datasets, one for season-by-season totals and one for career totals. For example, in the season-by-season dataset Saquon Barkley had five entries for each of his five seasons while in the career-totals dataset, Barkley had one entry with all of his career statistics. As we began to proceed, we noticed that many running backs had NAs (“Not Applicable”) listed for their statistics. In order to further investigate, we chose to examine Arian Foster, the 4x Pro-Bowl Houston Texans running back who had several slots in his statistic output listed as NA.
arianfoster <- pbp %>% filter(rusher_player_name == 'A.Foster') %>% select(rusher_player_name, rushing_yards, play_type, play_id, yards_gained, season)
For this piece of code, a filter was used so that the dataset would only include Foster’s rushes and give a play_id for each rush. A play_id is a nflfastR function that gives a unique identifier for every play
After assigning a play_id for each play and searching for NAs, the computer returned one NA, on a 2 yard run from Foster’s 2013 season. The description of that play was “TWO-POINT CONVERSION ATTEMPT. A.FOSTER RUSHES RIGHT TACKLE. ATTEMPT SUCCEEDED”. The computer had not been given a method for quantifying rushing yards for a two point conversion; there was an internal argument about whether it should be counted as two yards, or if it should be counted for any yards at all. Therefore, it was returning NAs for all 2PT conversions. We decided to assign all 2PT conversions as 0 yard rushes because these yards are not included in player or team rushing statistics.
With the NA issue solved, we needed a way to label any given season by a running back after a rookie year as a successive year in the league as a year in his career. For example, Saquon Barkley’s 2018 season was his rookie year, with 2019 as his second year in the league, and so on. This allows for us to graph the trends of a running back as they play more seasons in the NFL. Here is the result for Los Angeles Chargers running back Austin Ekeler:
Now that we have established years_in_league, we can finally get to graphing our data!
For our first graph, we examined how Rushing Yards change as a RB plays additional seasons in the NFL. This is the graph for all RBs from 2011-2021 with at least 100 carries per season:
There is a clear nonlinear trend between Years Played in the NFL and Rushing Yards for a running back. The blue line of our graph is a regression line that predicts the mean rushing yards for all running backs based on how many years they have played in the NFL. The grey region represents the 95% confidence interval of the blue regression Line. What this means is that we are 95% confident that the true mean rushing yards across all NFL running backs lies in this shaded region. So, as the shaded region increases, variability increases because the 95% confidence interval is larger. This regression line shows an increase in rushing yards from a running back's rookie season until their fourth season, followed by a steep, negative decline for the remainder of their career. The shape of this curve lends itself to the theory of a learning curve for NFL RBs followed by a “fatiguing phase”, injury phase, and as defenses figure out how to slow them down.
Rushing yards is not the only way to evaluate running backs, as it is very dependent on the number of attempts that a running back receives. Obviously, more touches would result in more opportunities to gain yards, so touches become a confounding variable in the analysis. The following graph suggests this to be true with an extremely strong correlation of 0.9688.
A different metric that we can use is YPC (Yards per Carry) which is defined as:
YPC = (total rush yards)/(total carries)
The graph relating seasons to YPC looks like:
With this graph, there is a clear negative relationship between the two variables. As a player plays more seasons, their Yards Per Carry strictly decreases. The steady decline in YPC tells a somewhat different story than the total yards displayed in the earlier graph. Here, we see that older RBs need more touches to produce the same amount of output. In other words, a running back becomes less and less efficient on a per-rush basis as they get older. This could be due to the toll that age takes on a running back’s speed, agility and durability. Many NFL teams have noticed this trend, and utilize RB "committees" to limit workloads to offset this expected decline in production. The average career length for any given NFL running back is 2.57 years, shorter than any other position player. The data suggests that the rushing yard peak in year three is supportive of this phenomena.
To better illustrate how this data plays out in real life, we chose to examine a specific NFL running back compared to our model. Every NFL season has different circumstances, so there are different variables every year that could impact a running back’s production. When picking a player to analyze, we examined the team they were playing for, competing players in the backfield, and injuries they had during their career. We eliminated any players who had had a major injury, defined by missing eight or more games in a season. We did not consider strength of schedule, changes in offensive personnel or scheme, and weather conditions. We then looked at players who had been drafted in the earlier years of our study, from 2011 to 2015. We chose Isaiah Crowell, who spent four years with the Cleveland Browns and one with the New York Jets.
In his rookie season, Crowell split carries with fellow rookie Terrance West, but in his second year he was the clear starter after West was traded away to the Tennessee Titans. In his next three seasons, Crowell performed somewhat consistently, but was cut before the 2018 season, shortly before future star running back Nick Chubb was drafted.
In his final season in the NFL, Crowell split carries with the Jets’ starter Bilal Powell, but led the team in nearly every rushing category. Crowell followed the model in terms of rushing yards, but his YPC did line up with the model.
Similar to the graph for all NFL running backs, Crowell rushed for more yards from his rookie year to his third year, reaching a peak of 952 rushing yards before experiencing the “running back wall”. However, the graph of Crowell’s yards per carry shows a very interesting trend:
In Crowell’s second season, his YPC dropped to 3.8, significantly worse than the league average of 4.4. One reason for this was that the 2015 Cleveland Browns, who finished 3-13, were constantly playing from behind and had to rely on a pass-heavy offense. Out of the 989 offensive plays run by the 2015 Cleveland Browns, a mere 38% of them were rushing attempts. Because he had so few opportunities to run the ball, his YPC dropped dramatically. In his third season, however, Crowell had a breakout season racking up a career high YPC of 4.8. Crowell’s efforts helped the 2016 Cleveland Browns finish second in the league in YPC (4.9 as a team). Unlike the composite YPC graph, Crowell’s YPC improved by the end of his career. Whereas the average running back’s YPC drops as he gets older, Crowell’s YPC was inconsistent and had no clear trend. In both year 3 and year 5, Crowell was part of a running back committee, allowing for a balanced workload and reducing injury risk and fatigue. This allowed for his YPC to defy the trend.
We wanted to understand why the value of the running back has drastically shifted from a team’s superstar, to more of an afterthought. In today’s NFL, the running back is falling in the draft, being paid less and replaced more often. The purpose of our study was to understand why this opinion has emerged, and to define the “running back wall.” Through the nflfastR package, we filtered play-by-play data from 2011 to 2021 and created regression models relating age to both rushing yards and yards per carry. Our visualization between years in the league and rushing yards showed that the average running back rushes for more yards from his rookie season to his third season, before sharply declining until the end of his career. This shape represents a learning curve followed by a “fatiguing phase,” as injuries take a toll on a running back’s health. Our visualization between years in the league and yards per carry showed a consistent decline as running backs play more years. As the age of a running back increases, they need to get more touches to have a similar output, making them less efficient and more replaceable. Our models support the idea that veteran running backs are not valuable due to the wall they hit prior to signing their second contract. Additionally, running back committees help limit workloads, allowing running backs to stay healthy and contribute to a team for longer.
By Field General Analytics Co-Founders Nate Yellin & Jack Gewanter
Kennedy, C. (2010, January 1). Age vs. wear and tear: The deterioration of an NFL running back. Institutional Scholarship Home. Retrieved June 24, 2022, from https://scholarship.tricolib.brynmawr.edu/handle/10066/6074
Friedlander, M. (2021, September 17). Mel's perspective: Don't give Najee Harris every snap. Steelers Depot. Retrieved June 24, 2022, from https://steelersdepot.com/2021/09/mels-perspective-dont-give-najee-harris-every-snap/
Average NFL height and weight by position - University of Idaho. (n.d.). Retrieved June 24, 2022, from https://webpages.uidaho.edu/~renaes/251/HON/Student%20PPTs/Avg%20NFL%20ht%20wt.pdf
Gough, C. (2019, September 10). NFL: Average career length. Statista. Retrieved June 24, 2022, from https://www.statista.com/statistics/240102/average-player-career-length-in-the-national-football-league/
2015 Cleveland Browns Team & Player Stats. StatMuse. (n.d.). Retrieved June 24, 2022, from https://www.statmuse.com/nfl/team/cleveland-browns-58/stats/2015
Team rushing averages in 2016. StatMuse. (n.d.). Retrieved June 24, 2022, from https://www.statmuse.com/nfl/ask/team-rushing-averages-in-2016