The (NFL passing) game done changed

I wrote last week about the Seattle Seahawks' offense changing, a shift that's mainly reflected in Russell Wilson's low air yards per attempt (AY/ATT) in 2014 (i.e., how far the ball is traveling in the air each attempt, or in other words, yards gained minus yards-after-the-catch). The Seattle attack is definitely tending further toward yards-after-catch (YAC), and, on an anecdotal level, this seems to be true league-wide.

This hypothesis came to the fore on Tuesday morning, when John Clayton spoke about the state of the 2014 passing game.

Clayton says SACKS and AIR YARDS are down big time NFL Wide from 2013....teams throwing shorter passes.

— DAVIS HSU (@DavisHsuSeattle) October 7, 2014

In an attempt to quantify these changes, I've used historical data from both Pro Football Reference and SportingCharts.com.

I've done my best to find reliable data for this study. It appears that the SportingCharts 2014 air yard data matches that of Advanced Football Metrics. Their historical air yard numbers diverge slightly from other yearly averages I've seen reported, but they're the only site I've found that publishes full values for years prior to 2014. The difference is marginal (within a few hundredths of a yard, generally), so it's my judgment that the data presented here is sufficient for our purposes.

Before we get too far, it's important to understand that air yards are the yards gained by a passing play prior to the catch. Every passing play is broken into two distinct categories: air yards and yards-after-catch. By definition, then, the two always add to 100% of the passing total. A 50-yard bomb into the end zone is 100% air yards, and a bubble screen at the LOS is 100% YAC.

Are Air Yards Actually Declining?

A little research on the topic of AY/YAC yielded a column from 2012, authored by Football Outsiders' Scott Kacsmar. I recommend reading the full article as he does an excellent job of studying the value of YAC and how it relates to quarterback performance. For our purposes, the money quote is the following:

With receivers like Danny Amendola, Percy Harvin and Tavon Austin taking on big roles, we may only be a few years away from a league where the lines between air and YAC intersect. -- Scott Kacsmar

Kacsmar is referring to the AY/ATT and YAC/ATT lines. Historically, the greater chunk of passing yardage has occurred prior to the catch. A few years down the line, we're not quite to an intersection, but it's very close. The short passing game has come into style in recent seasons, meaning there's an almost even split between yards gained before and after the catch. This is unprecedented.

Things were broadly consistent for most of the 90s; the peak in 2004 coincided with the introduction of illegal contact, and most other years have generally hovered around a 1 YPA difference between air yards and YAC. Between 2007 and 2009, this gap was halved, and it's plummeted even further since the start of 2013.

Now, AY/ATT in 2014 isn't all that different from the same parameter in 2001 or 2009. It's a little low this year, but it's not difficult to imagine that the number could slightly increase as the season progresses, coming more into line with the last five years. Basically, we don't have quite enough of a sample to confidently state that quarterbacks are throwing for significantly fewer air yards per attempt than in previous years.

While air yards are marginally lower, the the YAC increase is much more clear-cut. The following plot shows the percentage of total yards represented by both air yards and YAC. Again, a bubble screen would register at 100% YAC, and a bomb into the end zone would yield 0% YAC.

(Yes, these are mirror images.They're directly related and including both is redundant, but I've still presented both here as I feel it better emphasizes how close they are to an even split).

It's also interesting to view this phenomenon through the lens of individual quarterbacks, so I've included 24 individual QB career YAC% trends. Jay Cutler wasn't excluded because I'm anti-smoking, the plots are just already hard to read at 8 QBs each. There are a number of guys who didn't make the cut.

The following figures are dense and mainly exist to show just how much variance there is in each player's YAC% over a number of seasons. Players change schemes, have penalty-heavy years (drawing pass interference rather than completing deep passes), and face different schedules. It's not uncommon for a player to have a year where YAC is uncharacteristically high, like Steve Young in 1997.

Note that the x-axis is not zero here. I've compressed the axis range so that the trends are a little easier to pick out.

If you're interested in AY/ATT rather than YAC%, check here, here, and here. (For the brave of heart: all 24 QBs in the same (very) dense plot (higher-res, click to download). Godspeed.

The previous three plots are a bit of a mess. It's fairly rare for a QB's AY/YAC to be consistent from year-to-year, which gives us a little context for the dramatic change in Russell Wilson's 2014. To update last week's results, he now stands at a flat 3.00 AY/ATT (and over 60 YAC%) through four games. Note that the third Harvin non-TD (a deep pass) would've significantly changed this, so there are small sample effects still at play.

What does this mean? Wilson's throwing for far fewer air yards than in previous seasons, but it isn't all that anomalous. However, it does have a large (negative) impact on his performance in some of the advanced QB metrics, like QBR or DYAR.

Beyond Air Yards

There's a lot changing, and not all of it's captured in the AY/YAC issue. The following plot shows just how much of an outlier 2014 is thus far.

Well, hello, holding and illegal contact emphases. This season has produced historic marks in SACK% (sacks per dropback), INT% (interceptions per pass attempt), and TD% (touchdowns per pass attempt). Note that TD% is just a way to normalize touchdowns thrown by attempts, and a similar concept is at work with INT% and SACK%.

There's been a little bit of decline in yards-per-completion and a slight uptick in yards-per-attempt, but completion percentage and quarterback rating have seen more significant change over the last few decades. I'm using quarterback rating in this example as people are familiar with it, but there are better metrics available, like Adjusted Net Yards per Attempt (ANY/A). More on this later.

We see steady growth in both, but it's tough to fully understand just how extreme the change is from this visual representation. Though the blue increase doesn't seem huge, there's been a 6.2% gain in completion% since 1992, 4.2% of which came since 2005.

Passer rating was at 75.3 in 1992, near Chad Henne's career average of 75.5. In 2014, the league average of 87.7 is higher than Dan Marino's career 86.4. Of course, Dan Marino was a better player than the average 2014 quarterback. The point isn't that quarterbacks are better. The game done changed.

Takeaway: the league-average quarterback in 2014 is putting up numbers that would've made them a shoo-in to the 1992 Pro Bowl.

Passer Rating

As an aside, the look at completion% compared to passer rating was fairly interesting, as a common criticism has been that passer rating is too highly correlated to completion%.

The following plot represents normalized distributions of both passer rating and completion%, plotting z-scores of each against each other. This figure isn't imbued with a great amount of actual meaning (for a variety of reasons), but I found it visually interesting to see the apparent link between the two.

When season average passer rating is plotted as a function of season average completion%, the two metrics share an R-squared of 0.86. Now, we would expect a correlation between passer rating and one of its inputs, but it seems strong, perhaps overly so. This isn't to say there's zero value in passer rating, but it's important to understand that completion% is likely over-emphasized in its formulation.

In simple terms: passer rating will overvalue players with high comp% and low YPA (yards-per-attempt), while undervaluing those with low comp% and relatively high YPA.

Chase Stuart's ANY/A is my preferred quarterback metric. It's relatively simple and backed by clear methodology. The change from 1992's 4.9 ANY/A to the present 6.4 ANY/A is just as staggering as that shown by passer rating.

It's difficult to know if the extreme 2014 results are going to stay at their historic levels all season, but the new penalty emphases do present a good argument for the increase. It seems reasonable to expect at least part of the gains to stick. I hope to check in on this research again a bit later in the year when we have a little more data to parse.

---------------------------------------

Zach on Twitter