Brian Burke, ESPN Analytics 5y

We created better pass-rusher and pass-blocker stats: How they work

NFL

ESPN Analytics is pleased to present a revolutionary new way of measuring the pass-block and pass-rush performance of individual NFL players. Read the abbreviated explanation on how they work, or skip to the full details on why they matter and what they can tell us. And see what's new for 2019.

OK, what is this exactly?

Our new Pass Block Win Rate metric tells us the rate at which linemen can sustain their blocks for 2.5 seconds or longer. Likewise, our Pass Rush Win Rate metric tells us how often a pass-rusher is able to beat his block within 2.5 seconds. Our model of pass blocking harnesses player tracking data from NFL Next Gen Stats.

So, what can these stats help show?

Now we finally have objective individual stats for linemen for their most critical tasks -- defending and attacking the passer. We can also know who is blocking whom on every snap, who was double-teamed, who got pressure from the edges and who got pressure by collapsing the pocket. The end result is that we can assess individual performance and team-level performance in the trenches separate and apart from the performance of the quarterback, receivers and secondary.

How do they work?

It's actually pretty simple -- there's no fancy machine learning involved. Our model uses the location, proximity and orientation of each player relative to every other player throughout a play to determine who is blocking whom. When a pass-rusher beats his block, we can tell which blocker allowed the pressure. And just as importantly, we know how long after the snap it occurred.

Why are they better than stats?

Metrics like QB pressures and time in pocket might be useful, but they can be misleading. A QB pressure can occur for several reasons other than unreliable pass protection, such as good coverage, poor route-running or missed reads by the quarterback. Our win rate metric isolates line play from those other factors. Also, time in pocket metrics don't know the difference between a quick read and release by the quarterback and ineffective pass protection. If a passer throws at 1.8 seconds after the snap, does that mean he only had 1.8 seconds to throw, or did he execute his read quickly? Our metrics know the difference.

Where could these stats lead?

You might find these metrics appear on various ESPN shows and in our articles. When you see Pass Block Win Rate (PBWR) or Pass Rush Win Rate (PRWR) powered by Next Gen Stats, these metrics are what we're referring to. These stats offer a novel way of understanding what drives the success or failure of pass offenses and defenses. We can now objectively assess individual player performance quickly and accurately, at a scale and scope not possible before. We can marry our metrics to other advanced metrics such as expected points and win probability to directly measure the effect individual performance has on team success.

What's new in 2019 for PBWR and PRWR?

Our pass-blocking and pass-rushing metrics have been in use for more than a full season now, and we've learned a tremendous amount since we rolled out the initial version. This was one of the first major projects using player tracking in football, and much of our efforts are quite experimental. Although we were thrilled with the initial results from our metrics, we've made a number of significant improvements for 2019:

  • Improved handling of double-teams: We corrected a number of false positive rush-wins when a rusher beats one lineman in the double-team but not the other. Additionally, we were previously tallying only double-teams between two offensive linemen. Now we tally double-teams regardless of the position combination, so running backs and tight ends count too.

  • Improved handling of unusual pocket geometries: In an effort to capture when a rusher is able to collapse a pocket, we analyzed pocket geometry throughout each pass play. But our logic was often fooled whenever the pocket became "convex" (bending away from QB instead of around him), making it appear the rusher has caved the pocket when he hasn't. We've fixed this.

  • Corrected occasional false positive rush wins: Our algorithm depends on proximity and orientation to determine who is blocking whom. So when a blocker is slow to engage immediately after the snap, it can look like a rusher is unblocked. Similarly, if there is about to be a double-team and the nearer blocker is slow to engage, it could appear as though the rusher has instantly beaten the block from the more distant blocker. This also was fixed.

  • Significantly improved identification of which players are truly blocking and rushing: Our original logic merely defined rushers as defenders who crossed the line of scrimmage and defined blockers as offensive players who did not cross the line of scrimmage. This was acceptable for the vast majority of rushers and blockers, but wasn't very accurate for tight ends, backs and linebackers. The new logic is adapted from our latest player tracking project and is much more accurate. Block-and-go-type releases by tight ends and backs who do not cross the line of scrimmage are no longer counted against them as pass-block losses (and no longer count as pass-rush wins for the defender who is released).

  • Team-level PBWR and PRWR summary calculations now use the same logic as the individual-level metrics: We previously used a slightly different way of calculating the rates for teams than for individuals, but our improvements this year help lessen the need for different methods.

We can expect the improvements to cause PBWR numbers to increase and PRWR numbers to decrease as a whole. Double-team rate numbers will increase. The bottom line is that overall accuracy is substantially improved.


All the details on PBWR and PRWR

What they can do

Finally, an objective, meaningful stat for individual linemen and pass-rushers. Blocking and beating blocks are at the heart of football, but are among the most overlooked aspects of the sport. Pass Block Win Rate and Pass Rush Win Rate can tell us which linemen are having the biggest effect on their team's fortunes, and which are holding their team back. We can also now catalog every block on every pass play. We know who blocked whom and how long that block was sustained, which is the key to our metrics' effectiveness.

Just as a taste, below are the top defensive ends/outside linebacker rushers with more than 50 total rushes through Week 4 of 2018. The league average is 22 percent across those positions:

And here are the top offensive tackles in Pass Block Win Rate. The league average for tackles is 79 percent:

We can do the same kind of analysis at the team level, which can lead to some interesting insights. Take the Rams' resurgent offense in 2017, which more than doubled their points scored from 2016. A new scheme brought in by head coach Sean McVay is widely credited for the turnaround, which is hard to deny. But a large part of it was due to the arrival of tackle Andrew Whitworth and center John Sullivan in 2017. As a unit, the Rams' Pass Block Win Rate went from 35 percent in 2016 (30th in the league) to 55 percent in 2017 (second in the league). So far this season, Whitworth is the No. 2 overall tackle, just behind his teammate Rob Havenstein, and Sullivan is the No. 3 overall center. The Rams' offensive line deserves a lot of the credit for the team's turnaround, those two key additions in particular.

We can also see who is getting double-teamed the most. These tend to be the most dangerous pass-rushing defensive tackles, such as Aaron Donald, which makes his eye-watering Pass Rush Win Rate numbers all that more impressive.

How our metrics work

Our metrics rely on player tracking data from NFL Next Gen Stats, which offers complete data starting with the 2016 season. But we don't use the data to train a machine learner or anything like that. The amount of data allows us to apply the metrics to those plays, but doesn't necessarily enhance the accuracy/fidelity/etc.

The system watches every pass play, including scrambles, using the raw position and orientation data. It looks at which players are pass blocking and rushing and calculates a score based on the distance from every player to every other player along with the difference in orientation from each player toward every other. That score is used to determine who is blocking whom. Put simply, if you're at or behind the line of scrimmage, extremely close to a defender and facing him, you're probably trying to block him.

The system also draws the pocket at each time step in a pass play. The pocket is defined as a polygon connecting all the pass blockers. When a rusher penetrates this polygon and is within a certain distance to the quarterback, we say he has won his block. This is typically inside pressure, and prevents the passer from stepping forward into passes or away from edge pressure.

Block wins on the edges and in other situations are determined when a rusher becomes closer to the quarterback than who is blocking him and he is within a certain distance. We exclude long distances, because there are often cases where a QB has rolled out far in one direction, and beating a block is inconsequential to the play.

We exclude all screen plays, because the intent is often to allow pass-rushers into the pocket. Check out the animation below, in which each step is 0.1 seconds of real time: 

When blocks are held for at least 2.5 seconds or more, we call that a win for the blocker. When blocks aren't sustained for at least 2.5 seconds, we call that a win for the rusher. We chose 2.5 seconds for two reasons. First, it's about the average time to pass release on most conventional drop-back pass plays. It also appears to best separate the blockers and rushers we already know are good at what they do -- blockers such as David Bakhtiari and rushers such as Joey Bosa, who were both ranked No. 1 in their respective win rates for their positions in 2017.

We calculate these metrics on both the individual player level and the team level. One important difference between the two versions is that unblocked rushers who threaten the quarterback, who are usually late blitzers, are counted at the team level and counted for the rusher himself. However, because there is no blocker to debit for the free blitzer, it is not counted against any individual blocker.

It happens that the average team-level win rate on each pass play is about 50 percent in the last couple seasons. Individual Pass Block Win Rate across all blocking positions is about 75 percent, and the individual Pass Rush Win Rate metric is close to 25 percent.

Why this is a major improvement over existing measures

Although metrics such as QB Pressures and Time In Pocket can be useful, they are often misunderstood and don't tell a complete story.

QB Pressures are a binary metric, and don't tell us how long it took to generate pressure, which can tell us what and who the cause of the pressure is. They don't tell us who allowed the pressure, or whether someone had to defeat a double team to create it. The metric is also dependent on the quarterback's reaction. Some quarterbacks will stand in the pocket and either deliver the throw or take a hit, while others will scramble away. Those who are qualitatively determining whether a pressure existed on each play might have varying interpretations and definitions influenced by how a quarterback reacts. Our new metrics are totally objective and (with some minor exceptions) independent of the quarterback's reaction.

An example of what this can show: In 2017 the Chargers ranked third in team Pass Rush Win Rate, but in our own pass pressure rate metric we had them ranked 22nd. I think most people who followed the Chargers' defense in 2017 would agree they were significantly better than 22nd in that department.

Time In Pocket (TIP) numbers are misunderstood. They're easy to measure but hard to interpret correctly. You start the stopwatch at the snap, and then stop it when either the quarterback releases a pass or is sacked. (Some versions also stop the clock when he leaves the tackle box.) A team can have a low TIP because they make quick reads and release the ball right away, or they could be giving up a lot of pressure early in the play. A team's high TIP could be due to good blocking, or it could be due to a quarterback that can't make reads (like the Browns in 2017) or a receiving corps that can't get open (like the Ravens in 2017).

Perhaps the best example of the flaws of TIP metrics: The 2017 Steelers, who ranked 22nd in TIP but had one of the most high-powered, deep-threat passing offenses in memory. Despite Ben Roethlisberger's early-career reputation, the Steelers tended to make quick, decisive reads and got the ball to receivers who got open quickly.

Pass Block Win Rate says a big part of the reason their passing game was so effective was that they were ranked No. 1 overall in pass protection that season.

Pass Block Win Rate and Pass Rush Win Rate solve the flaws of Pressures and TIP stats because they properly analyze pass blocking from the perspective of systems engineering. Yes, systems engineering. Pass protection is best understood as a parallel process, meaning if any one link in the chain breaks, the entire chain fails. That is, if any one blocker allows pressure, then the quarterback is indeed pressured no matter how well the other blockers do.

Time to failure is an essential measure. Those in industrial or military pursuits may be familiar with a metric called Mean Time Between Failures (MTBF), which tells you how reliable a factory machine or weapon system is before it breaks. Our approach is similar. We look at pass protection and pass rush as what's known as a "survival function," which conveys how likely a system is to last beyond any given time. Here's what a survival function would look like as we compare the Rams' Pass Block Win Rate in 2016 and 2017. The higher the curve, the longer the blocks are sustained.

There is one obvious limitation of our new metrics-adjustment for opponents and double-teams. Blocking a top pass-rusher is obviously a difficult task compared to double-teaming against a replacement-level rusher. Things like play action and rollouts also change the difficulty of blocks and rushes.

Repeating these metrics for run plays is also high on the to-do list. We started with pass plays, because they are more critical to game outcomes than run plays and, honestly, they're easier to analyze. Assignments are clear, and the objective is always to protect the quarterback as long as possible or sack the quarterback as quickly as possible. Run plays tend to be more complex, chaotic and generally aren't concerned with time from snap. But it's just a matter of time before we're able to analyze run blocks as well as pass blocks.

For more from ESPN Analytics, visit the ESPN Analytics Index.

^ Back to Top ^