So that we might better comprehend the future outlook of NFL players, over the next couple of months before the NFL season, I plan to look into aging curves for every position in the NFL.
I previously looked into the aging curve for defensive players as a whole, and was surprised to find that defensive players peak at age 23 and, more importantly, are close to their peak from the very young age of 22 when most players enter the league.
When I went to look into the specific aging curve for each position, my data, or lack thereof, became a problem.
I had used Pro Football Focus (PFF) overall grades to conduct the original research for the aging curve of defensive players, and since Pro Football Focus overall grades are only available to the public for the years from 2007 to 2014, it was impossible to be able to come up with any statistically significant conclusions on the position specific level.
It’s impossible to come up with an aging curve for players without a value metric to judge them by. Sure you can use yards-per-carry for running backs or passing touchdowns for quarterbacks, but those metrics don’t account for the total value that a player provides, only certain aspects of their value. To top it off, positions like tackle, guard, and center are completely void of any common metric that would lend to the overall assessment of a players worth.
How we’ll make do…
To substitute for Pro Football Focus overall grades, we will use Pro Football Reference’s approximate value (AV) as our performance metric to judge a player’s ability. Approximate value, as quoted on Sports Reference, is “an attempt to put a single number on the seasonal value of a player at any position from any year (since 1950).”
There are a couple of problems that I have with AV. Unlike PFF overall grades, AV does not capture what a player was able to accomplish independent of their team (e.g. one of the component’s of AV includes team points scored). This means that great players on bad teams are penalized because of the ability of the other players on their team.
Another issue that I have with AV is the emphasis that is placed on accolades. Players are given credit for first-team all pro, second-team all pro, and pro bowl selections. This is unfortunate in my mind, because if a player receives an award, that doesn’t mean they should have received an award (e.g. Ryan Clady made the pro bowl last year and he had a negative PFF overall grade and ranked as the 41st best tackle of players that played on 25% or more of their team’s snaps in 2014; Jahri Evans made the Pro Bowl last season, which is laughable because of his ineptitude in pass protection).
With that said, approximate value is better than a sharp stick in the eye. Without it, we have nothing, and with it we at least have a general frame of reference for the value a player produced in a given season. It doesn’t pretend to be perfect; Sports Reference even tells you this when they say “AV is not mean to be a be-all end-all metric.” It is just supposed to be approximate.
With AV as our value metric, we will be able to shine some light on a topic that, to this point, has been barren of robust methodology and research, and come up with an idea of the general path that players at each position take as they mature and devolve throughout their careers.
This series will use the same methodology that was used for previous research that I did on the aging curve for defensive players.
The delta method is used to minimize the inherent survivorship bias that plagues attempts at aging curves in general. There is still an intrinsic survivorship bias in this method, but it is smaller in this method than in any other methodology known to the author at this point
Mitchell Lichtman gives a thorough explanation of the delta method in this series of posts: Part 1 and Part 2. However, the methodology used in our study mirrors the work that Jeff Zimmerman did in this research.
If you read the post on the aging curve for defensive players, you can skip the rest of this section and go to straight to the section that talks about the population.
This is the detailed explanation on the delta method from the post I wrote on the aging curve for defensive players. This explanation uses PFF overall grades as the value metric, but, remember, we plan use AV for this project:
The most common mistake that people make when it comes to aging curves is that they conflate two separate issues: how a player played and how a player aged.
Suppose that you looked at the average grade for middle linebackers that are 29-years-old and 30-years-old. The 29-year-olds have an average grade of +6, and the 30-year-olds have an average grade of +10.
For starters, the 30-year-old linebackers are the linebackers that survived until their age 30 season. In order to survive in the NFL until your age 30 season, you have to be a talented player. Because these 30-year-olds have been able to play for so long and have the players at the bottom of their population filtered out over time, this means that we have a selection bias in the sense that the group of players that we’ve decided to look at are inherently talented because of their ability to survive.
As the current group of 30-year-old middle linebackers moved from their age 29 season to their current age 30 season, they lost players at the bottom of their population. These lost players are the linebackers that were too slow and/or weak to continue to play. This same group of players that could no longer continue to play is still present in the current group of 29-year-old middle linebackers, while they are absent from the current 30-year-old group of linebackers; this group at the bottom deflates the average grade of the 29-year-old players and inflates the average grade of the 30-year-old players.
Because we know that the 30-year-old linebacker’s higher Overall Grade average is a product of a selection bias, it doesn’t really make sense to look at the average of player seasons. The current group of 30-year-old linebackers, on average, is more talented than the current group of 29-year-old linebackers, but that doesn’t mean that they 30-year-olds weren’t more talented as a group the year before in the previous season—their age 29 season.
Sure, there may be a player here and there that had a better age 30 season than age 29 season, but, as a whole, the current population of 30 year old players will have played better the previous year when they were 29.
The Delta Method
Now that we know that it doesn’t make much sense to look at the average Overall Grade of players in a specific age bracket, we can shift our focus toward the analysis of how a player’s performance changes over time.
This research use the delta method, which is summarized in this article by Jeff Zimmerman, who helped me throughout this process and whom I owe a big thank you to. Thank you.
The delta method looks at change over time. For our purposes, the delta method looks at the change in specific player performance from year-to-year and weighs that difference in year-to-year performance by the harmonic mean of the player’s snap totals for the two years that were used to find the difference.
Lets use Luke Kuechly as an example. Kuechly had a grade of +28.4 in 1,015 snaps as a 23-year-old inside linebacker. As a 22-year-old, Kuechly had an overall grade of +11.1 in 1,012 snaps as an inside linebacker.
Under the delta method, we would follow these steps with Kuechly’s stats:
Year 1: Overall Grade = +11.1 in 1,012 snaps at age 22
Year 2: Overall Grade = +28.4 in 1,015 snaps at age 23
Calculate the harmonic mean for snaps: 2/((1/Snaps_y1)+(1/Snaps_y2))=Snaps_hm
2/((1/1,012)+(1/1,015)) = 1,013.49
Weight the Overall Grade to the Snaps_hm:
(Snaps_hm/Snaps_y1) * Overall Grade_y1 = Overall Grade_y1_hm
(1,013.49/1,012) * +11.1 = +11.1 Overall Grade for Year 1
(1,013.49/1,015) * +28.4 = +28.3 Overall Grade for Year 2
Kuechly gained +17.2 in his Overall Grade over 1,013.49 snaps from age 22 to 23.
Because the number of snaps played were so similar in each season, the weight of the harmonic mean in the case of Kuechly’s age 22 and 23 seasons did not make much of a difference, but a measurable difference can be seen when players have a large variance in snaps from one season to the next.
After this has been done for all players and seasons, the differences in Overall Grade and snaps are added together and then adjusted to 500 snaps.
Football Reference provides AV numbers for all players since 1950, however, since players most likely don’t age the same way today that they did in 1950, we will use a more recent time frame of 1980 to 2014 for our population of players. All players and their AV from these years will be used to determine the rate at which players age for each position.
For this first post, as you can glean from the title, we will look at the aging curve of offensive tackles today.
There are 2,445 NFL players that played offensive tackle in consecutive seasons from 1980 to 2014 that are in this specific population.
The graph below shows the aging curve for offensive tackles from 1980 to 2014.
(This graph and the conclusions reflect changes that were made on 6.24.15; there was an error found in the initial curve. The information in the graph and the conclusions is now up to date. Thanks, Devin.)
What we can see:
- Offensive tackles peak at age 28.
- In their first few seasons, tackles take three huge steps forward with their performance.
- After those first big steps, players steadily increase in performance until their age 28 season.
- After players reach their peak at 28, their performance falls off somewhat and they age gradually until their age 31 season, at which point performance falls off more rapidly in their age 32 season.
- The peak period for an offensive tackle seems to be from their age 25 to age 31 season.
The goal of this project is to generate a deeper knowledge about the common course, or lack thereof, of future performance for NFL players.
Specifically for what was just shown for offensive tackles, what we have learned is that offensive tackles seem to be a wise long-term investment for NFL teams: shocker. They have a steady rise to their peak, and after a slight drop-off, age nicely into their early-30s.
To contrast this to what we’ve already learned, while defensive players are essentially at or close to their pinnacle when they enter the league, offensive tackles need a much longer time to mature.
As a result, teams need to set realistic expectations when it comes to the development of their players, and, for example, not hold an offensive tackle to the same standard of development that defensive players are held to.
More importantly, from a roster composition perspective, it seems like it makes sense to give offensive tackles long-term contacts, while it seems like the same cannot be said for defensive players. Offensive tackles peak after their rookie contracts expire and retain a large amount of their abilities into their early-30s, while defensive players are already on their decline after their rookie contracts expire.
As of now, we can only speak to how defensive players age collectively, but, as this project goes on, we will be able to find out it if only a certain group of defensive positions peak at the start of their careers or if this is a symptom of the collective.
This project will hopefully reveal a more intelligent way to construct an NFL roster—does it makes sense to commit to certain players long term or at certain points in a player’s career and not others?
I plan to write next about the aging curve for NFL running backs, and the work that I have done on this so far might surprise a lot of people. Follow me on twitter to get updates about the project and be notified of the newest articles; hopefully we can find a lot of valuable insight along the way and come up with a more objective way to look at the overall outlook of NFL players of all positions as they advance throughout their tumultuous careers.
Photo Credit: Erik Drost
- Week 13 NFL Breakdown: Joe Flacco Shouldn’t Make You Think Twice - December 7, 2016
- Week 12 NFL Breakdown: Mark Ingram Rolls Against the Rams - November 30, 2016
- Week 11 Breakdown: DeVante Parker Could Be Turning the Corner - November 23, 2016