In this week's Varsity Numbers, Bill Connelly revisits some measures and concepts: Adjusted Scores, Covariance, and momentum (or whatever you choose to call it).
24 Jan 2013
by Danny Tuccitto
Remember how absolutely ancient the Pittsburgh Steelers defense was heading into 2011? If not, here's an NFL Network segment about it. Alternatively, you can dust off -- better yet, purchase! -- Football Outsiders Almanac 2011 to read about the oldest defense this century. Buried in our analysis was this sentence: "The planned lineup going into 2010 averaged 30.5 years of age, although the actual lineup for the season ended up averaging 29.5 years because Ziggy Hood replaced an injured Aaron Smith." The implication of that sentence is where we'll end our journey today, but let's start with a little bit of background on how to measure a team's average age.
In the past, calculations and discussions of which teams were too old or too young occurred -- and still occur -- during the "sweeps week" of average-age programming just prior to Week 1. Some report average ages based on rosters as they exist after final preseason cuts, while others, like Football Outsiders and ESPN's Mike Sando focus on the ages of projected starters. The problem with both of these approaches, of course, is that they're employed before any teams have actually taken the field during the regular season. As alluded to above, rosters and starting lineups are dynamic, so if the goal is to convey the relative youth of a team, then an ideal measure of average age should be dynamic as well.
FO also calculates average team ages after the season ends, but it's just another case of straightforward math based on starters at the end of the season. A more "advanced" after-the-season approach is Chase Stuart's method of weighting ages according to Pro Football Reference's Approximate Value (AV) system. The main advantage of weighting ages by AV is that it accounts for the common sense fact that some players are more important to their teams than others. For instance, 22-year-old starting quarterback Robert Griffin (14 AV) should have a larger impact on Washington's average age than 32-year-old third-stringer Rex Grossman (0 AV).
Although it's certainly an improvement over simple math, Stuart's AV-weighted averages still have a couple of drawbacks. First, AV gives no value to specialists, so kickers, punters, long snappers, and the hundreds of players who toil in relative obscurity on coverage units don't contribute to their teams' average ages. Second, because AV itself isn't calculated until after All-Pro teams are announced, AV-weighted ages, by definition, can't be determined until after the season is over.
Such was the situation until 2012. As part of an ESPN Insider column last month on Atlanta's shrinking window of championship opportunity (just as relevant six weeks later!), I created a new measure of average team age that's weighted, not by AV, but by each player's weekly snap counts, which the NFL decided to release publicly for the first time this season.
The logic of weighting by snap counts is simple: the extent to which a player's age matters is a function of how often he actually sees the field. It solves the problem of faulty assumptions like rosters or starting lineups remaining static through the season, it accounts for the fact that specialists actually do play, and it can be calculated at any point in time.
The method is simple as well. As an example, Ryan Tannehill played 62 of Miami's 1,914 total player snaps in Week 13, so his age (24) comprises 3.2 percent of the team's average snap-weighted age that week. For the season, Tannehill was on the field on 980 of the 28,619 snaps amassed by Dolphins players, so his age contribution to Miami's 2012 snap-weighted average is 3.4 percent. You might think these weights are tiny, but remember that a 45-man active roster on game day means each player, if weighted equally, would contribute 2.2 percent to average age in a given week. Furthermore, equal player weights would be even smaller for full-year average ages when you consider a typical NFL team in 2012 had 66 different individuals on their active roster over the course of the season.
The table below shows average snap-weighted ages for every NFL team overall, as well as those for its offense, defense, and special teams.
I'll return to the merits of the measure shortly, but a few quick observations are in order. Leaguewide, the average snap-weighted age of an NFL player in 2012 was just below 27 years old, with the average special teams snap being taken by a player nearly a full year younger than one on offense and defense. In addition, the four regular season games since my Insider piece weren't enough to dethrone Atlanta as the oldest team and oldest offense, Pittsburgh as the oldest defense, and Detroit as the oldest special teams.
Only three teams ranked among the 10 oldest across all team units: Atlanta, Baltimore, and San Diego. Congratulations, Chargers fans, your favorite team is both bad and old! On the other end of the spectrum, Cleveland and St. Louis had bottom 10 rankings across the board, which provides the former with an excuse for a bad season and the latter with an additional argument for why they're an up-and-coming team in the NFC West. Too bad division-mate Seattle was No. 1 in DVOA and had a younger snap-weighted age.
To my eyes, though, the biggest surprise in the table is Detroit's offense, which had the fourth-highest snap-weighted age despite Matthew Stafford (24) playing 98 percent of their offensive snaps, Mikel LeShoure (24) playing 44 percent, and Titus Young (23) playing 50 percent before he mentally imploded. The reason: their offensive line. Outside of first-round pick Riley Reiff (23), who ended up playing only 325 offensive snaps all season, no other Lions lineman who saw the field was younger than 28 years old: Jeff Backus (35), Dominic Raiola (34), Stephen Peterman (30), Rob Sims (29), and Gosder Cherilus (28) combined for 5,865 offensive snaps. Indeed, the snap-weighted age of Detroit's offensive line (30.5) was over a half-year older than the second-ranked New York Giants (30.0).
Getting back to the methodological improvements represented by snap-weighted age, you'll recall I listed them as actually accounting for (1) special teams and (2) roster dynamics. Let's take these one at a time.
It was a surprise to learn that Detroit had the oldest offense this season, but with Jason Hanson (42) as their kicker, and the combination of Nick Harris (34) and Ben Graham (39) as their punter, I'm sure no one was surprised that they had the oldest special teams. However, it's not that simple. Yes, if you just count kickers and punters (and perhaps long snappers) as the only "specialists," then it's a no-brainer. But what if I told you that Hanson, Harris, Graham, and Don Muhlbach (31) combined to play only 8.4 percent of the Lions' individual special teams snaps in 2012 (461 of 5,499)? It turns out that Kassim Osgood (32), Stefan Logan (31), and John Wendling (29) played nearly twice as many (896).
For reasons of player development and salary cap management, special teams by and large are the realm of an NFL team's young depth. Therefore, seeing a situation like Detroit's might be an indicator that the team in question doesn't have much young depth. After all, if you can avoid the $2 million cap hit associated with a pair of old, one-trick ponies like Osgood (five offensive snaps) and Logan (26 offensive snaps), you do so. And indeed, if you look at the rankings for snap-weighted age on special teams, you see the likes of Detroit, San Diego, Washington, Arizona, and Oakland (i.e., teams not associated with talented young reserves) among the oldest.
Contrast this with a team like the Packers, which had eight players under the age of 27 who played more than 20 percent of Green Bay's special teams snaps and at least 40 percent of their offensive or defensive snaps: Randall Cobb (22), Jerel Worthy (22), Casey Hayward (23), Jerron McMillian (23), Dezman Moses (23), M.D. Jennings (24), Sam Shields (25), and Brad Jones (26). Of particular note vis-à-vis Detroit, these eight players, of which all but McMillian started at least four games, had a combined cap hit of $4.5 million, with none having an individual cap number above $750,000.
So we've established that, compared to previous approaches, snap weighting is the only comprehensive measure of average age on special teams. That's obviously a good thing, and it's especially fortuitous given that the other snap-weighted ages (offense, defense, and overall) had astronomically high correlations with the traditional measures:
|Week 1 Rosters||.875||--||--|
|Week 1 Starters||.922||.895||.872|
|Week 17 PFR Rosters||.887||.757||.725|
|Week 17 Ourlads Starters||.870||.912||.864|
AV-weighted age for 2012 exhibiting high correlations with snap-weighted age is actually pretty intuitive when you think about it. In a nutshell, AV is basically an elaborate -- some might say genius -- way to quantify "participation level" in so far as it relies on yardage totals, games started, All-Pro selections, and the like, which by definition can't be achieved without a player seeing the field. But what is a snap count if not an even more explicit accounting of "participation level?"
Nevertheless, this similarity doesn't render useless our snap-weighted ages. As mentioned earlier, AV doesn't come out until after the season ends, and Stuart didn't do a post on AV-weighted ages for 2011 until August of 2012. Granted, that probably had something to do with his excellent site Football Perspective not launching until last summer, but either way we're talking about some amount of delay.
Snap-weighted ages, on the other hand, can be calculated at any time, provided you have the snap count data. (Which you do, because it's posted right here on FO weekly.) And where the fun really begins is when you plot how average snap-weighted ages change over the course of the season. For instance, here's a graph showing the trajectory of snap-weighted ages leaguewide (age on the vertical axis, week of the season on the horizontal axis):
Here we see that, overall, younger and younger players were on the field as the season progressed, with the league average dropping by about 0.5 years from its high in Week 2 to its low in Week 17. The situation was basically the same for offense and defense. However, this wasn't at all the case for special teams, which had an average snap-weighted age that barely moved all year. (And if you're wondering, that precipitous overall drop in Week 7 was due to offenses in Atlanta, San Diego, and Denver enjoying their bye weeks. The fact that the needle moved so much in their absence tells you just how old they were relative to the rest of the league.)
The first thing that comes to mind when seeing the steady decline of snap-weighted age over the course of 2012 is injuries. As more and more starters succumbed to the wear and tear of an NFL season, they were replaced with their younger backups. This theory seems to pan out when we look at individual teams. Below are the snap-weighted age graphs for the two participants in Super Bowl XLVII.
That's quite a contrast. Whereas the 49ers chugged along right around their snap-weighted average age (27.2) until Justin Smith got hurt, the Ravens got a full year-and-a-half younger between their high in Week 6 and their low in Week 17. Losing Ray Lewis (37) against Dallas, and replacing him with Jameel McClain (27) made the team on the field a week later half-a-year younger despite the return of Terrell Suggs (30). An even larger decrease occurred in Week 10, when DeAngelo Tyson (23) and James Ihedigbo (29) played the majority of snaps in place of injured veterans Haloti Ngata (28) and Ed Reed (34). Finally, after returning to baseline for a few games, Baltimore's snap-weighted age fell by a full year in the last two weeks thanks to players like (the other) Chris Johnson (33) and Bobbie Williams (36) going back to the bench after brief stints as replacement starters, as well as the Ravens retreating into their turtle shell for a relatively meaningless finale.
Although it sure looks like Baltimore's late-season slump may have been linked to getting younger, with only one season of data at our disposal, it's too early to tell if changes in snap-weighted age produce changes in game DVOAs, and it appears that age (or experience) is more of a mediating variable in the relationship between injuries and game performance: Older starters get hurt, thereby forcing younger, less experienced players onto the field, which in turn decreases performance efficiency. We'll certainly revisit this in the future. At the full-season level, though, preliminary evidence shows that the correlation between snap-weighted age and DVOA is largest on offense (.303). Sample size caveats apply, but older offenses were more efficient in 2012, probably due to the positive impact of older quarterbacks (e.g., Denver, New Orleans, and New England) and older offensive lines (e.g., Atlanta, Detroit, and the New York Giants).
As one final demonstration of how roster dynamics affect team ages, let's return to the Pittsburgh Steelers defense. As you'll recall, due to injuries, this unit shed one full year off their 2010 average over the course of the season, a development Aaron noted in FOA 2011. Well, two years later, the unit they fielded in Week 1 was actually 0.4 years younger than the one they fielded in Week 17. But the interesting part isn't where they started or where they finished; it's how they got from Point A to Point B.
Like most teams and most units, the Steelers defense got much younger towards the end of the year, mainly because of injuries to Ike Taylor (32) and LaMarr Woodley (28). And like Baltimore, their increased youth on the field tracked alongside a late-season slump. Atypically, however, Pittsburgh's defense experienced a rise in snap-weighted age of nearly two full years from their low in Week 1 to their high in Week 10. Of course, that was because of players like James Harrison (34) and Ryan Clark (33) returning from injury.
In essence, the effect of roster dynamics works both ways. It's not enough to simply account for it by taking snap shots of the starting lineup in Week 1 and Week 17. I've shown you three team graphs, and all three look incredibly different. As the season progresses, teams are impacted by injuries at different times, or in the case of, say, the Arizona Cardinals, you start the year with Kevin Kolb (28) at quarterback, then spend the middle part of the season starting John Skelton (24) and Ryan Lindley (23), but finish the year nearly right back where you started age-wise (27-year old Brian Hoyer).
As indicators of a moment in time, previous approaches may have ended up being accurate, but they weren't very precise. An NFL team isn't a static entity during the fall and winter, so a measure of its age should not be one either. On FO, and in almost all football media, we discuss a team's changing nature all the time. What is their weighted DVOA? Does momentum mean anything? We also talk an awful lot about special teams. Why, then, should we ignore these things when talking about age? Thanks to the NFL's release of snap count data, we no longer have to.
37 comments, Last at 26 Jan 2013, 1:01pm by bucko