Writers of Pro Football Prospectus 2008

Most Recent FO Features


» Week 3 Quick Reads

We break down the very ugly game Joe Flacco had against Jacksonville, how completely ineffectual the Ravens offense was -- and why it wasn't even the worst game among quarterbacks this week.

12 Jul 2017

2016 Play-Action Offense

by Bryan Knowles

We continue our pre-FOA 2017-launch statistics bonanza with a look at how offenses used play-action passing in 2016.

In 2015, the rate of play-action usage dipped to just under 19 percent after holding steady around 21 percent in the previous three seasons. That rate actually dropped a little bit more in 2016. It's down to just 18.5 percent now, the lowest since 2010. This is not exactly the most precipitous drop-off in the world, but it is consistent, and it brings us back to the sort of environment we saw in the '00s (the 2000s? The naughts? The Ottos? Whatever we're calling that decade now). After a three-year blip, we seem to be back to "normal" now.

It's important to note that there was a data-entry change between 2014 and 2015; we moved from using ESPN Stats & Info charting for play-action to using SIS charting with ESPN data as a check. Any time you see a significant change in data immediately after changing how you assemble said data, it's worth being at least a little suspicious. There are enough plays where you look at a little feint and wonder "was that actually a play-fake?" that judgment calls come in along the margins, and at least some of the difference could easily be one set of charters being more liberal in what they call play-action than others. These are some of the fun, exciting things you experience in the world of subjective data entry.

If the drop was consistent among all teams, that's exactly what I would call it: a data glitch; a statistical footnote. But the drop-off isn't consistent across the league. What we're seeing, in large part, is the end of the zone-read era.

Six teams had at least two seasons in the 2012-2014 "hot zone" where they ran play-action on more than a quarter of their passes -- Washington, Seattle, Philadelphia, Carolina, Minnesota, and San Francisco -- and all six have fallen back towards the mean in recent years. That's pretty much a who's-who of zone-reads and read-options. Robert Griffin, Russell Wilson, Cam Newton, and Colin Kaepernick were the "big four" dual-threat quarterbacks to enter the league in 2011 and 2012. Chip Kelly brought his Oregon offense with him to the pros in Philadelphia. Minnesota's something of the odd team out, but even they were in the illustrious Christian Ponder era.

Those six teams used play-action on more than 30 percent of their passes in 2012 and 2013, thanks in large part to the sheer number of zone-read plays they brought to the table. While other teams may have seen their play-action rate fluctuate, the consistency of the Big Six seems to be what was driving the league-wide increase in play-action rate.

Play-Action and the Big Six
Year League PA % Big Six PA % Other 26 PA % Difference
2011 19% 23% 18% +5%
2012 21% 33% 19% +14%
2013 21% 30% 19% +11%
2014 21% 26% 20% +6%
2015 19% 24% 18% +6%
2016 18% 19% 18% +1%

Of course, neither Kelly, Ponder, Griffin, nor Kaepernick currently have NFL jobs -- take that much rushing philosophy away from the league, and it's no surprise that play-action rates have fallen off of a cliff. The Panthers and Seahawks have also dialed their play-action down in recent years, for a number of reasons. Their quarterbacks have continued to mature and develop as passers. Their offensive lines have grown, if anything, more questionable at holding their blocks.

Most importantly, though, is the fact that defenses around the league have learned how to cover read-option plays, with scrape exchanges and double A-gap blitzes and the like. Teams still use the read-option, but they've also moved on to using more run/pass options and package plays for their misdirection. It's not like the zone-read was a Wildcat-esque fad, but it's no longer the zeitgeist of offensive football like it was a few years ago.

Right, so that covers the drop-off. But what about the play-action plays that actually were run in 2016?

  • Standard deviation dropped off yet again in 2016, down to 3.0 percent (from 3.8 percent in 2015 and 5.6 percent in 2014), so the Big Six coming back to the pack hasn't been replaced by new teams seeing their usage rise dramatically.
  • DVOA on play-action plays (including passes, sacks, and scrambles) rose to 27.8%, up from around 24.5% last season -- and, indeed, the best success rate we have ever charted. This is despite non-play-action passing DVOA actually dropping from 14.0% to 10.7%, so it's not like teams just suddenly became a lot better at passing last season. Play-action passes will always be a little more successful than your run-of-the-mill passes because of their situational usage -- they tend to be called on non-obvious passing downs, and lead to deeper throws on average -- but the reason the gap increased specifically last season isn't immediately clear. It could just be a one-year blip in the data, as there's a lot of noise from year to year.
  • Yards per play-action pass remained steady at 7.9, compared to 6.2 for non-play-action passes. Deeper throws lead to more yards per attempt, shockingly.

The following table lists play-action data for the 2016 regular season only. Offenses are sorted by descending rate of play-action usage (as a percentage of dropbacks). The sections next to usage list results on all play-action plays (including scrambles), play-action passes only, and normal passes. The final section shows the difference between standard play-action and normal passes. We subtracted normal pass DVOA from play-action pass DVOA, so a high ranking in the difference column indicates the offense performed better with play-action.

Play-Action Usage, Offenses, 2016
Offense PA Pct Rank With PA (Pass/Scram) With PA (Pass Only) No PA Difference
Yds/Play DVOA Rank Yds/Play DVOA Rank Yds/Play DVOA Rank Yds/Play DVOA Rank
ATL 27% 1 10.3 53.3% 5 10.5 53.4% 5 7.8 50.5% 1 1.8 2.8% 25
CIN 24% 2 7.5 27.2% 18 7.7 26.0% 18 6.7 22.0% 10 0.8 5.3% 22
DAL 24% 3 8.8 45.3% 9 9.0 45.8% 7 7.0 40.7% 4 1.8 4.6% 23
SF 23% 4 7.1 18.2% 20 7.1 16.6% 20 5.8 -4.9% 26 1.3 23.1% 12
MIA 22% 5 7.3 40.1% 12 7.0 37.9% 12 7.4 8.6% 15 -0.1 31.5% 11
TB 22% 6 8.1 67.0% 2 7.9 66.0% 2 6.3 7.2% 18 1.8 59.8% 3
BUF 21% 7 6.9 28.5% 16 6.9 28.0% 17 6.1 11.0% 14 0.8 17.5% 14
SEA 20% 8 8.7 40.6% 11 8.5 41.8% 11 6.7 8.3% 16 2.0 32.3% 10
CHI 20% 9 7.7 10.3% 26 8.1 12.0% 23 6.8 7.2% 19 0.9 3.2% 24
IND 20% 10 9.1 28.3% 17 8.8 28.5% 16 6.6 13.8% 13 2.4 14.5% 16
NE 20% 11 9.2 58.8% 3 9.3 63.0% 3 7.2 45.6% 2 2.0 13.2% 18
PHI 19% 12 8.3 46.1% 8 8.2 43.9% 9 5.4 -7.6% 27 3.0 53.7% 4
MIN 19% 13 9.2 50.1% 6 9.2 51.6% 6 6.0 -1.0% 24 3.2 51.0% 5
TEN 18% 14 8.3 20.6% 19 7.9 16.7% 19 6.6 31.1% 8 1.6 -10.4% 26
BAL 18% 15 5.7 -26.8% 31 5.4 -25.3% 31 6.2 7.3% 17 -0.5 -34.1% 32
CAR 18% 16 6.0 -9.0% 30 6.1 -9.0% 30 6.6 5.0% 21 -0.6 -14.0% 28
Offense PA Pct Rank With PA (Pass/Scram) With PA (Pass Only) No PA Difference
Yds/Play DVOA Rank Yds/Play DVOA Rank Yds/Play DVOA Rank Yds/Play DVOA Rank
DEN 18% 17 6.5 14.4% 23 6.6 13.8% 21 6.2 -0.1% 23 0.3 14.4% 17
WAS 18% 18 10.4 47.7% 7 10.7 45.6% 8 7.2 31.8% 7 3.2 15.9% 15
CLE 17% 19 6.2 7.0% 28 6.3 6.5% 28 5.8 -13.4% 30 0.4 20.4% 13
KC 17% 20 9.2 58.5% 4 9.1 58.1% 4 6.2 15.0% 12 2.9 43.5% 7
HOU 17% 21 6.3 -40.2% 32 5.9 -41.4% 32 5.4 -10.2% 28 0.9 -29.9% 31
NO 17% 22 7.8 37.5% 13 8.0 37.5% 13 7.4 30.2% 9 0.4 7.3% 21
SD 17% 23 9.6 83.9% 1 9.7 85.6% 1 6.5 -1.5% 25 3.1 85.5% 1
DET 16% 24 6.7 9.4% 27 6.7 8.4% 26 6.8 20.5% 11 -0.2 -11.1% 27
NYJ 16% 25 8.0 6.9% 29 7.6 7.3% 27 5.8 -30.5% 31 2.2 37.4% 8
LARM 16% 26 8.3 15.6% 21 8.1 11.6% 24 5.1 -48.9% 32 3.2 64.5% 2
OAK 15% 27 6.6 42.4% 10 7.3 42.6% 10 6.5 33.6% 6 0.1 8.8% 19
GB 15% 28 6.4 11.8% 25 6.9 5.8% 29 6.9 41.2% 3 -0.5 -29.5% 30
NYG 15% 29 7.6 33.0% 15 7.5 33.0% 15 6.2 0.4% 22 1.4 32.5% 9
JAC 15% 30 6.6 15.3% 22 6.5 13.1% 22 5.8 6.8% 20 0.8 8.5% 20
ARI 15% 31 8.6 35.1% 14 8.6 35.1% 14 6.0 -11.6% 29 2.5 46.7% 6
PIT 14% 32 8.3 12.7% 24 8.3 10.4% 25 6.9 33.7% 5 1.4 -21.0% 29
NFL 18%
7.8 28.3%
7.9 27.8%
6.2 10.7%
1.6 17.7%

You can track Kyle Shanahan's movement throughout the NFL by tracking spikes in play-action usage. His Atlanta offense led the league in play-action rate at 27 percent, and that's not a fluke. Shanahan's offenses have ranked in the top 10 in every season since he took over as coordinator in Houston back in 2008, and this is the second time his offense has led the league (after Washington in 2012 in RGIII's first year).

Kyle Shanahan's Play-Action Passing
Year Team Play-Action % Rank Play-Action DVOA Rank
2008 HOU 22% 7 77.3% 3
2009 HOU 21% 7 75.2% 3
2010 WAS 27% 8 -2.1% 26
2011 WAS 21% 10 17.6% 18
2012 WAS 42% 1 66.7% 1
2013 WAS 27% 6 -13.0% 30
2014 CLE 31% 2 32.7% 14
2015 ATL 22% 8 30.3% 10
2016 ATL 27% 1 53.3% 7

Washington, Cleveland, and Atlanta saw their play-action pass attempts go up by an average of seven percentage points when Shanahan came rolling into town. With Brian Hoyer in Cleveland, Shanahan hit 31 percent back in 2014, so don't be shocked at all if San Francisco is topping the table here next season. Atlanta will likely drop back towards the league average as Steve Sarkisian takes over play-calling duties.

Shanahan used play-action more in 2016 than he did in 2015, but Atlanta wasn't the biggest riser -- that was Dallas, who shot from 15 percent up to 24 percent last year. Obviously, the addition of Ezekiel Elliott helps out tremendously here; defenses had to respect the running game in Dallas in a way they really haven't since Marion Barber's heyday, and that opens up to the threat of the play-action pass. Dak Prescott was very effective completing those play-action passes, but then, he was very effective whenever he was asked to throw the ball. It's questionable how much play-action passing really helped Prescott; the Cowboys ranked just 23rd in difference between non-PA DVOA and PA DVOA. Still, don't knock it if it works, right?

Pittsburgh ranked dead-last in play-action pass attempts, which isn't new -- they were next to last in 2015. This is in direct response to the fact that Ben Roethlisberger isn't actually very good at play-action passing. At all. The Steelers have had a negative DVOA difference in three of the last four years -- and, again, play-action passing will generally score higher in DVOA. It's not like Roethlisberger has trouble with the deep ball in general or anything like that, but all his stats fall off of a cliff when he's asked to fake handoffs. Very strange and, apart from a blip in 2014, very consistent. Todd Haley's teams in Kansas City and Arizona were generally much better when using play-action, so this is squarely on Big Ben's big shoulders.

Carolina saw the biggest drop-off in play-action rate, falling from 27 percent to 18 percent, and that makes a lot of sense. In 2015, Carolina had the second highest play-action pass rate in the league, but put up a -28.3% DVOA difference. They still ranked 30th in passing DVOA out of play-action last season, so that could easily drop further next year, but Carolina is rather radically redesigning its offense. We'll have to wait and see how the addition of someone like Christian McCaffrey alters their passing game plans.

One team that should probably use play-action passing more? The San Diego Los Angeles Chargers. For two years running, they had been dead-last in play-action passing rate. Frank Reich seemed to be allergic to the very concept of the play-action game, but the return of Ken Whisenhunt brought some much-needed variety to the Chargers' passing attack. Specifically, there seems to have been a conscious decision to stick Philip Rivers under center more, which opened up more room for traditional play-action passes. The result? Not only did they increase their volume (from 10 percent to 17 percent), but they also saw a huge uptick in their play-action DVOA (from 27.3% to 83.9%). This is now the third time in the last four years that the Chargers have finished in the top two in play-action DVOA. It's a perfect fit for Rivers' game and the Chargers' multi-tiered passing offense. The fact that Reich ignored it for so long had to have been maddening for Chargers fans.

The biggest drop-off in play-action success from 2015 goes to Houston, who fell from a respectable 17.0% DVOA to a league-worst -40.2%. Brian Hoyer has consistently been a good play-action passer -- we saw it in Chicago in 2016, in Houston in 2015, and in Cleveland in 2014. It's really the one thing he does well on a regular basis, and the 2015 Texans took advantage. Both Brock Osweiler and Tom Savage, on the other hand, were terrible last season. More to the point, they were even more terrible at play-action passing than they were on non-play-action passes, somehow. The current golden standard for grading bad quarterback play is "were you worse than Jared Goff?" and yes, yes they were: Goff "only" put up a -41.0% DVOA on play-action passing, better than either Osweiler (-41.1%) or Savage (-43.5%). Please win a starting job soon, Deshaun Watson.

Before ending, I want to follow up on something we touched on last year. Generally, the link between good play-action offenses and normal dropbacks is high. Before 2015, the correlation between DVOA in the two categories usually fell somewhere between 0.35 and 0.70 -- a fairly solid correlation, when you take into account individual quarterbacks' strengths and weaknesses. That plummeted to 0.16 in 2015, with some very strange rankings. Did that continue, or was it a one-year blip?

Correlation in 2016 did in fact rise -- but only to 0.29. That's still significantly below what we saw going back through a decade of tracking this, and that's rather strange. Last year, we suggested that the low correlation might be a one-year blip, or that the increasing number of packaged plays were asking for a significantly different skill set than traditional dropback passes. These could both still be accurate. The teams that used play-action more often than normal -- say, 20 percent of the time as opposed to the league average of 18 percent -- saw a correlation of 0.53, so maybe those teams doing a better job of making play-action part of their offensive identity are seeing more consistency between their various DVOA ratings? That feels like cherry picking to me.

Perhaps we're just entering an era where some quarterbacks are simply better at running packaged plays and other non-traditional sets than they are at straight dropback passing. But then again, is there really a common thread between Rivers, Jameis Winston, Alex Smith, Sam Bradford, and Carson Wentz, all of whom put up higher than 45.0% DVOA on play-action and lower than 15.0% otherwise? Conversely, is there something that links Roethlisberer, Aaron Rodgers, Jay Cutler, Matthew Stafford, Cam Newton, Joe Flacco, Brock Osweiler, and Tom Savage, all of whom were below 15.0% DVOA on play-action and put up better numbers when they work in more traditional passing situations? It seems like two fairly random groupings of quarterbacks to me, with very little in the way of rhyme or reason.

Sometimes, weird splits and weird correlations happen without obvious reasons for them -- it's something I'll be keeping an eye on this year, but I'm currently at a loss for explanations. Sometimes data is just weird.

Later this week, we'll flip over to the other side of the line and check out how defenses fared against play-action.

Posted by: Bryan Knowles on 12 Jul 2017

9 comments, Last at 29 Aug 2017, 5:20pm by msabercr


by jacksonmcg :: Wed, 07/12/2017 - 8:10pm

One thing Rodgers, Cutler, Stafford, Newton, Flacco, and Roethlisberger have in common is arm strength. I don't know how true this is for Osweiler or Savage but their combine profiles mention this as a strength for them too.

Why these two things would be related though I'm not sure and I don't think that the other set is considered a set of weak-armed QBs.

by Bryan Knowles :: Thu, 07/13/2017 - 12:35am

Good observation! I'm with you -- I'm not quite sure how it would match up (maybe strong-armed QBs don't need as much play-action misdirection to go deep?), but that's a pretty decent arm strength list, all things considered.

As for the other side...

Smith certainly is a weak-armed QB, and despite Bradford's arm strength, he relied mostly on shorter passes last season, as did Wentz -- but not Winston.

by Aaron Brooks Go... :: Thu, 07/13/2017 - 9:56am

Wentz's receivers couldn't catch deep passes, so they stopped throwing them.

by MilkmanDanimal :: Thu, 07/13/2017 - 10:28am

A large portion of Tampa's offense last year consisted of Winston flinging the ball downfield in the vague direction of Mike Evans, and Evans then doing something amazing. If you combine the utter lack of other downfield threats with the fact the Bucs started I believe four different RBs and the Bucs' success on play action is just weird. I mean, they weren't good at running the ball and it's not like the offensive line was spectacular in general, so not sure why play-action was that much more successful than a standard offense.

by PirateFreedom :: Thu, 07/13/2017 - 9:23am

Interesting observation.
maybe play action is more likely to succeed as misdirection when defenses don't fear the QB's arm as much.
maybe having a great arm makes QBs less dependent on perfecting play action fakes to stay in the league.

by msabercr :: Tue, 08/29/2017 - 5:20pm

My guess is PA thins the field giving QBs an easier time at identifying separation. So new QBs, in a new scheme, with new teammates that don't have that rapport with their receiving corps, and haven't had success throwing deep, have a better chance of stretching the field with play action and being successful. In contrast all the people that already have an established relationship like Ben, like Stafford, like Cam have no major hang ups identifying separation and are comfortable letting the ball loose on those deep looks because that's their major skill set; Identifying the coverage and making quick snap decisions on favorable match ups deep.

by Darren :: Wed, 07/12/2017 - 11:12pm

I think Atlanta's PA vs. no-PA difference is 2.9, not 25.9; assuming the other columns are correct.

by Bryan Knowles :: Thu, 07/13/2017 - 12:22am

2.8, actually -- it's comparing the difference with scrambles included -- but nice catch. I've edited the table to match the correct numbers.

by Scott C :: Thu, 07/13/2017 - 3:12pm

Rivers has a strong arm and is accurate at distance.

And that explanation doesn't cover the other side of the equation at all. It tries to answer why some quarterbacks get more benefit from play action, but it does not explain why some are _worse_ at it.

I think the scheme and QB's ability to sell the fake is a more likely explanation. Or ability to take their eyes off the field and and re-locate WRs and read defenses after the fake.

EDIT: this was supposed to be a reply to the first post, but somehow ended up below.