Stat Analysis
Advanced analytics on player and team performance

2018 Rate of Adjusted Pressure

Deshaun Watson
Photo: USA Today Sports Images

Guest column by Ajit Kirpekar

Welcome to the 2018 Rate of Adjusted Pressure Numbers! And yes, just in time for the 2019 upcoming season which is upon us. For those who missed the origins and methodology behind RAP, see this link. Reminder, all data comes from Football Outsiders' game charting data (collected first by FO staff and volunteers, then in conjunction with ESPN Stats & Info, and since 2015 by Sports Info Solutions).

For those of you who made it all the way through my last article, a big salute to you! Not only did you read all about my tireless efforts, but hopefully you got a nice little taste of neural networks. Those who missed that article or do not want to go through that part (and I certainly don't blame you) can skip to the bottom where I posted the results.

Before I get into last year's rankings, I want to highlight three technical changes I made this time around. First, I enhanced the opponent rating system by adding in injury data as well as weather adjustments (this for the opponent ratings, not for the pressure model -- those were already there). Now a team's rating will vary from week to week based on its injury count and the game conditions. This did improve the model but caused a shift in the all-time rankings.

The second technical change was that I improved my text search process. Pressure is marked by game charters, but there is a field for additional comments where charters will remark if a pressure was missed or misreported. I was able to extract that information as well. This also caused a shift in the all-time rankings.

The final technical change was to the model itself. I'm still using the same overall framework, but I added a few additional bells and whistles which I'll save for the comment section if people want to hear about them. Suffice to say, technology in this area changes pretty fast, and new and improved methods come around quickly.

OK, enough about the boring stuff. Let's visit 2018.

Below is a table with two teams listed. As a fun little exercise, can you guess which team is which?

Team Defensive RAP Rank
Team A 32.89% 6
Team B 20.89% 32

The big offseason story from a year ago was the blockbuster Khalil Mack trade. Enough literal and digital ink was spilled about it, including a wave of articles about the winner and loser. Even the analytics community weighed in. Leaving aside who "won" the trade, it sure seems to have had an impact. I won't declare it definitive -- certainly a lot of other things changed for both teams beyond Khalil Mack's arrival/departure -- but you'd be hard pressed to find anyone who thought the Raiders' pass rush would get better from this. And the Raiders are Team B, ranking dead stinking last in RAP. As for Chicago, they finished 14th in RAP in 2017, although the year-to-year improvement was only about 2 percentage points. That may suggest, at the very least, that Mack helped stave off regression.

From the biggest headline of last offseason to the biggest of this offseason, I thought it might be interesting to look at Andrew Luck and his tenure with the Colts. Below are the Offensive RAP's for each year where Luck started for the majority of the season.

Year Offensive RAP
2012 38.59%
2013 33.51%
2014 27.17%
2016 37.22%
2018 26.76%

In 2018, Luck had the lowest Offensive RAP of his career. By contrast, most of his career has seen numbers almost ten percentage points higher. Again, it's spurious to come to any definitive conclusions; it is after all one data point, and football is a complicated sport. However, it does at least hint at the possibility of how much improvement can happen over a year even with a quarterback who has historically been a pressure-prone player. We all wish Luck well on his next endeavors, but seeing such a market improvement does inspire some what-ifs.

OK, onto the full rankings. Below are the 2018 Offensive RAP Rankings.

Offense Pressure Rate RAP Difference
NO 23.73% 19.77% +3.96%
PIT 23.21% 22.03% +1.18%
NE 24.71% 22.90% +1.81%
LAR 26.82% 23.23% +3.58%
OAK 27.15% 25.58% +1.57%
CIN 27.62% 26.08% +1.54%
IND 27.29% 26.76% +0.52%
ATL 27.94% 27.44% +0.50%
PHI 28.22% 27.72% +0.50%
DET 28.47% 27.80% +0.68%
CAR 28.98% 28.18% +0.80%
CLE 29.27% 28.34% +0.93%
CHI 28.85% 28.36% +0.49%
BAL 29.95% 28.41% +1.53%
TEN 27.97% 28.79% -0.82%
NYG 29.17% 29.99% -0.82%
LAC 32.03% 30.56% +1.47%
GB 31.34% 30.93% +0.41%
NYJ 30.86% 30.93% -0.07%
TB 33.63% 32.50% +1.13%
WAS 32.59% 32.68% -0.09%
KC 35.17% 32.72% +2.45%
MIN 32.38% 33.36% -0.98%
SF 33.34% 33.52% -0.19%
DAL 34.22% 33.55% +0.67%
DEN 34.14% 33.79% +0.35%
JAX 34.13% 34.69% -0.56%
ARI 34.74% 34.87% -0.14%
MIA 36.83% 36.72% +0.12%
SEA 37.48% 37.59% -0.11%
BUF 35.35% 38.53% -3.19%
HOU 40.96% 41.58% -0.62%

FO's own Rivers McCown shared the following Tweet after Luck's retirement:

I can imagine Texans fans seeing Deshaun Watson running for his life as a scary harbinger for a future we all want to see avoided. Apportioning blame is a complicated business, but Luck's marked improvement in RAP provides a nice data point that it can indeed be done. On the other hand, seeing yet another Seahawks team languish at the bottom of the rankings has become an annual tradition. To stress home the point, this was the Seahawks best Offensive RAP with Russell Wilson as the primary starter, and it ranked as the 26th worst Offensive RAP team recorded since 2006. That's right, this was their best finish, meaing that every single one of his other teams has finished in the bottom 25, with multiple in the bottom five. In other news, due to the technical changes made above, the 2014 Seahawks are now only the second-worst Offensive RAP team in the data set. Replacing them for that dubious honor are the 2007 Steelers.

On the flip side, the best performing teams were, unsurprisingly, teams that were either very good at passing the football or notorious for throwing short to avoid pressure. I added that proviso because while you see teams like the Rams and Saints are high up the list, Kansas City and the Football Clippers are only middle-of-the-pack, while Oakland ranks quite high despite fielding an average pass offense by DVOA. There are multiple interpretations you can make, but seeing this list reminded me of a common saying that coaches and pundits echo all the time: "Don't turn the ball over and you win!" I don't think even they believe that, because if they did, then the offense would return back to the dark ages of running up the middle and throwing passes to running backs. Clearly, coaches realize that turnovers are a tradeoff of risk-taking, and I think yielding pressure/sacks is yet another tradeoff you make when it comes to offense. Kansas City is probably happy to trade some additional pressures for a chance at a big play. This seems to have defined Aaron Rodgers' and Russell Wilson's careers.

I do want to make one additional point: if there's an inverse to the Seahawks, it's the Drew Brees-led New Orleans Saints. Aside from a few hiccups, every single one of his teams rank highly in Offensive RAP, and thus an unsurprising first-place finish in 2018. I mentioned Peyton Manning's ability to avoid pressure in my previous article, but Brees deserves similar recognition.

OK, onto the defensive side.

Defense Pressure Rate RAP Difference
LAR 38.09% 36.75% -1.34%
PIT 36.39% 36.44% +0.05%
BAL 33.01% 34.20% +1.19%
BUF 31.96% 33.50% +1.54%
NYG 32.12% 32.91% +0.79%
CHI 34.16% 32.89% -1.27%
CAR 31.50% 32.41% +0.91%
MIN 32.49% 32.21% -0.28%
WAS 31.00% 31.93% +0.93%
DAL 32.47% 31.89% -0.58%
KC 32.15% 31.12% -1.02%
GB 32.88% 31.02% -1.86%
HOU 30.18% 30.97% +0.79%
JAX 31.60% 30.85% -0.75%
TEN 30.84% 30.65% -0.19%
NO 31.41% 30.59% -0.82%
MIA 29.35% 30.48% +1.13%
CIN 28.03% 30.45% +2.42%
NE 32.25% 30.20% -2.05%
IND 30.86% 30.09% -0.77%
SF 30.76% 29.78% -0.97%
PHI 30.64% 29.77% -0.87%
DEN 31.03% 29.48% -1.55%
CLE 28.24% 29.12% +0.89%
LAC 31.33% 28.91% -2.41%
SEA 29.94% 28.85% -1.09%
DET 29.11% 28.60% -0.51%
NYJ 27.85% 28.31% +0.46%
ARI 29.25% 28.20% -1.05%
TB 28.18% 28.09% -0.09%
ATL 26.16% 26.70% +0.55%
OAK 22.80% 20.89% -1.91%

As mentioned above, the Raiders finishing at the bottom was a somewhat predictable result, so there's no need to linger on that issue. Let's start at the top with the L.A. Rams. I have called Aaron Donald a destroyer of worlds. It's hard to call him underrated at this point, given his numerous accolades, but the rams finishing first in Defensive RAP is one more well-deserved tip of the cap.

It's interesting that both the Rams and Steelers finished extremely highly in Offensive RAP and Defensive RAP. RAP isn't everything, but such strong finishes do drive home the point that opportunities for special seasons are rare; when you don't win, it just adds to the laundry list of what-ifs. In Pittsburgh's case, it's hard to fathom a team this good managing to miss the playoffs ... but hope springs eternal Steelers fans.

I do want to spend a few words on the defending champs. A look over their history provides an interesting window into weird dichotomies. Believe it or not, all three teams from 2006 to 2008 finished in the top 15 in Defensive RAP. From 2009 to 2011, the Patriots' Defensive RAP was in the bottom 50 before slowly rising to consistent mediocrity. Every year since the middle of this decade has seen more or less the same performance. In the prior article, I mentioned the murkiness of assessing pass rush as a function of overall pass defense, but even I am surprised at how successful the Patriots have been while fielding such mediocre units. Year after year, pass-rushers are highly paid and highly drafted, yet somehow the Patriots manage without them. "In Bill We Trust" I guess.

 

Ajit Kirpekar is a data scientist at Snowflake based out of San Francisco, California. Despite this, he roots for the Indianapolis Colts. You can follow him on Twitter @akirp.

Comments

7 comments, Last at 05 Sep 2019, 2:08pm

1 The majority of pass plays occur without pressure;

That's why the Patriots can get by without a superior edge rush; you devote the resources to cover well, or to trick the opposing quarterback into a turnover. That's why the 2009 Jets had such a dominant defense even though they collected only 30 sacks; let Revis cover the best receiver and devote the rest of their secondary (which was excellent) take out all the other options. Rex Ryan's blitzes helped, but you know, Belichick has tricks up his sleeve as well. Since 2010 New England has had Devin McCourty back there as either a corner (his rookie year) or safety, and he's a big reason they keep turning the ball over.

2 Economics are undoubtedly…

Economics are undoubtedly another part of that.  I'm sure the Pats would love to have a top edge rusher and a higher pressure rate, but there's only two ways to get one:

a) with a high draft choice in round one, and the Pats routinely draft at the bottom of the pack each round, or

b) by paying a veteran, and paying them well, which leaves less money available to invest elsewhere.

The benefits of pass rushers are well understood throughout the league.  I'm not quite sure how NE keeps winning, but it isn't by trying to find value in a space everyone else is already looking.

3 Fair but I think everybody…

Fair but I think everybody would argue pass-rushers are pretty integral. No matter how value seeking you are, if something is essential, then it's essential. And the Patriots seem to be completely alone in this regard. 

 

In fact I tried to find another team in the data that managed to be so successful over a long period of time well Fielding below-average pass-rushing units. I didn't find one.

4 It Bears Investigating

The Patriots have relied, especially in recent years, on complementary football, a good offense and special teams, generally forcing opponents to work on long fields. Also exploring market inefficiencies and bringing in mid-level free agents while creating overall roster depth. But that's not all. They tend to give away a fair amount of underneath stuff and play very disciplined to avoid big plays, requiring opponents to put together long play sequences. Once the opponent falls behind - as often happens against Brady - and tries to force things, turnovers happen. There was a long discussion on Football Perspectives on the Pats and the myth of 'bend don't break'. And BDB is a myth, except with the Pats, apparently, who have given up far fewer points than their defensive rating would predict. I do think that BB started to alter his philosophy over the past two or so years in response to an aging Brady. There is more attacking and pressure, and apparently greater investment in talent on that side of the ball, especially in the defensive backfield.  

5 Well, I hope this doesn't…

Well, I hope this doesn't come across as crapping on this article, but I am trying to get a handle on the utility of RAP. It seems to be an overly complicated process to derive a very small improvement over the unadjusted, naive pressure rate. The r^2 between RAP and pressure rate is about 0.95 (for 2018 offense) and I don't see any significant outliers, which could be useful information. So, what does RAP tells us that regular old pressure rate does not? With some problems (e.g., healthcare applications), the use of complicated models to incrementally improve a key metric is well worth the effort. I just don't see that here unless I am missing something.

6 Short answer...because of my…

Short answer...because of my snappy commentary? 

 

Longish Answer: The R2 for 2018 between the naive pressure rate and RAP is 0.95 (It drops to 91 percent for defensive btw), but historically, it has been about 70 percent for the full sample.  

In addition, the year to year correlation in pressure rate is about 10 percent lower than in RAP, suggesting RAP is picking up something that the naive pressure rate is missing.

Furthermore, consider a situation where a team scored 300 pressures in 600 snaps, but rap suggests a team actually should have gotten 330 pressures to 600 snaps. That amounts to a seemingly trivial 5 percentage point difference in pressure rate, but the 30 additional pressures has some deeper significance.  30 additional pressures is like two games worth of additional pressures(and maybe even 10 more sacks for the team itself). Consider how much team's pay for 30 additional pressures and 10 sacks. 

 

As for complicated model - I am not sure which part you find complicated - the model itself or the opponent adjustment methodology?

 

If its the former, well - that just happens to be what the best fitting model was. I tried a number of other things, an ordinary logistic regression, random forest, xgboost, etc etc - its what came out the best performing. 

 

For the latter - I will admit, its a bit of a philosophical exercise and by nature its complicated. I could also adjust for weather, how many first down attempts a defense had to face vs third and longs, field types, etc etc. Adding those does create much larger swings than the ordinary version - but this gets into a philosophical discussion. Should we be punishing teams who happened to play in doors or offenses that avoid a lot of third and longs? Teams definitely dont walk onto a stadium blind to the weather conditions or the opponents they face - they augment their behavior(something the model cannot understand). So I had to ultimately make some personal decisions on how far to take the adjustments.

 

All that to say, the year to year correlation between dvoa and my elo adjustments(using point spreads), is something like 85 percent as well. So why bother with Dvoa, which relies on parsing play results, turnovers, injuries, weather, etc etc,  while my (previous) version needs only point spreads and can be run on a laptop in milliseconds? Because we want have the most accurate measure of team performance we can get. Sure, squeezing that extra 10 percent is nowhere near as valuable as the medical field or self driving cars, but its still hard won and tells us something. 

7 Thanks for the thorough…

Thanks for the thorough answer and additional insights, especially interested to hear the correlation between naive pressure rate and RAP is usually quite a bit lower. That suggests that some teams RAP can deviate substantially from NPR, much more than the above tables would indicate. This changes my whole outlook on the ROI of such an approach.