# Party in the Backfield

By Austin Tymins

As the NFL gets more and more pass-reliant, having an effective pass rush has become a necessity. In this post, Iâ€™m going to look at the 2014 season in detail to identify the best pass rushers and best sacking teams. To do this, it is important to incorporate an expected points model so as to compare the relative value of a sack to other plays that occur in a football game.

For those unfamiliar with expected points, it essentially an answer to the question: which team is most likely to score next and how many points will they score? Expected points models are created using a Markov chain, regression, or recursive function. In this study, Iâ€™m using a smoothed variation of the Advanced Football Analytics expected points model. Additionally, the data itself comes from the NFL Savant play-by-play database for 2014. I removed all sacks in which there was no play, meaning Illegal Contact, Defensive Offsides, or Defensive Holding and credited half sacks between the two players involved.

After building the expected points model, I was able to find that the average sack in the NFL lost 6.36 yards and 1.24 expected points for the offense. However, this doesnâ€™t come close to representing the true value of a sack. If a team starts a drive 1st and 10 from their own 20-yard line, they have an expected point value of .34 points. If a 6-yard sack occurs, they will be 2nd and 16 from the 14-yard line, equivalent to an expected point value of -.73. In this case, the sack was worth 1.07 points for the defense. However consider the opposing example where the offense is 3rd and goal from the 1-yard line and has an expected point value of 5.31. A 6-yard sack results in 4thÂ and goal from the 7-yard line and an expected value of 2.6 meaning the defensive sack is worth 2.71 points. Clearly the value of a 6 yard sack is not constant across down and field position.

With this framework in place, we can start to analyze all the sacks that occurred in the 2014 season. The sack with the greatest expected point value (most â€śvaluableâ€ť sack) came from Cameron Wake on Tom Brady on September 7, 2014. The Patriots were looking at 4th and 10 from their own 18-yard line before being sacked for 8 yards. Therefore Wake was credited with 3.55 â€śsackâ€ť points (sack EPA). The Wake sack essentially clinched the game and was even more impactful from a win expectancy viewpoint; however, we are only looking at expected points and are thus ignoring high leverage game states in this post.

The histograms of sack yards lost and sack EPA for the 2014 season are below.

As we can see, there are a few cases in which a sack has produced a negative EPA. These typically occur when a team is starting from a very difficult late-down situation deep in their own territory. When a low yardage sack occurs, the team avoids the turnovers deep in their own zone that often lead to easy points and are forced to punt the ball away.

The next histogram shows the histogram of all players that produced sack EPA last season. As we can see, Justin Houston and J.J. Watt appear to be extreme outliers. The table of the top 15 players in sack EPA is also below.

The sack EPA ratings are actually distinctive from the simple sack count rankings. For example, DeMarcus Ware shot up from 15th in total sacks to 4th in Sack EPA. Cameron Wake rose from 11th to 6th. Everson Griffen wasnâ€™t as fortunate and fell from 9th in the total sack rankings out of the top 15 as measured by sack EPA. The next table shows the average sack EPA top 15. Some unexpected names appear on this list, although that is partially because I didnâ€™t set a baseline number of sacks to be considered.

Once I saw these rankings I began wondering: are better pass rushers more likely to produce valuable sacks? Are the sacks J.J. Watt produces better than the regular NFL sack on average? Below is the scatter of average sack EPA by total sack EPA restricting for at least 3 sacks so as to curb some of the simultaneous causality. In the regression output just below that, we see that having a higher total sack EPA does in fact lead to better sacks at a statistically significant level.

Now that weâ€™ve analyzed individual sack performance, it makes sense to turn our attention to the team side. The method for calculating team sack EPA is exactly the same as in the individual player case. The team sack EPA rankings are below.

It shouldnâ€™t surprise many to see the Baltimore Ravens atop the rankings considering their perennially top-notch pass rush was tied for 2nd in traditional sack rankings. At number 4, the Jaguars are an unpredictably effective sack defense, and Cincinnati finished last mostly due to a slew of injuries to the front seven. Another interesting finding is that we now have a value for one component of the pass rush. In a future post, I will attempt to quantify the value of QB hits, hurries, and pass deflections to find a total EPA for pass rush.

Our regression using last yearâ€™s fitted values didnâ€™t return a significant result or explain much variation in team Elo ratings. However, the lack of statistical significance is likely due to the small sample size of 32 observations. The coefficient of .775 is actually relatively large in a real world sense. Holding everything else constant, having the NFL-best Baltimore pass rush instead of the Cincinnati rush is worth 29.7 Elo rating points. This is approximately the difference between the 2014 Atlanta Falcons and New York Jets. The formula for converting Elo ratings into win probability (Pr(A) = 1 / (10^(-ELODIFF/400) + 1)) shows us that an elite pass rush, holding everything else constant, adds 4.6% win probability.