Though lacking in glitz and glamour, the tackle, guard, and center positions make up the backbone of every NFL offense. Without skilled players in those roles — and players who can work as a unit — a team’s entire strategy can fall apart. In the past few years, teams like the Rams, Chiefs, and Saints have used a punishing offensive line to ignite high-powered offenses, while the Patriots have revolutionized O-line versatility. Even while these once anonymous units are finally getting their due, new analytics measuring offensive line performance just might prove that we’re still underrating these guys. In this week’s interview, ESPN’s Seth Walter discusses the growing field of O-line analytics, and just how much winning the battle of the trenches correlates with winning the battle of the scoreboard.
Paul Michelman: First, we’d like to ask you a favor. If you enjoy our program, please take two minutes to rate us — and even better — post a review on Apple Podcasts or wherever you get Counterpoints. In fact, you can hit pause and do it right now, if you are so inclined.
Super Bowl 48 was built to be the meeting of an unstoppable force — the Denver Broncos’ high-octane offense, led by Peyton Manning — and an immovable object: the Seattle Seahawks and their shutdown secondary, the Legion of Boom. But while that matchup headlined the marquee, there was a reason the Broncos were favored heading into the game: a mismatch on the offensive line. Working behind a group that gave up the fewest sacks in the NFL, Manning had the time and protection to shatter the records for passing yards and touchdowns in a single season.
Meanwhile, his counterpart Russell Wilson had succeeded in spite of an O-line that was dead last in adjusted sack rate and had given up six sacks in the playoffs already. Seattle’s historic defense had overcome these woes to reach the big game, but it looked like O-line play would be this Super Bowl’s scale tipper. And then exactly one play into the game, the script was flipped. Denver center Manny Ramirez snapped the ball over Manning’s head and into the end zone for a safety, and from that point on, the vaunted Denver offensive line wilted. Though only sacked once, Manning was routinely pressured in the pocket. He threw two interceptions and Denver had a big old goose egg on the scoreboard for almost three quarters. And that worrisome Seattle line? They went sack free for only the second time all season, allowing Wilson to control the game from the backfield and giving the Broncos no chance to get back in it.
I’m Paul Michelman and this is Counterpoints, the sports analytics podcast from MIT Sloan Management Review. Ben says he’s too busy teaching this week to join the show. I question his priorities. In any case, in this episode, we’re taking a look at those unsung heroes of the gridiron, the offensive line, and how new analytics are demonstrating just how closely tied O-line performance is to winning.
Though lacking in glitz and glamour, the tackle, guard, and center positions make up the backbone of every NFL offense. Without skilled players in these roles — and players who can work in tandem with each other — the team’s entire strategy will fall apart. In just the past few years, teams like the Rams, Chiefs, and Saints have used a punishing offensive line to drive high-powered offenses both on the ground and through the air, while the Patriots have revolutionized O-line versatility on their way to more than a handful of rings. Increasingly, these once anonymous units are getting their due. Even so, new analytics measuring offensive line performance just might prove that we’re still underrating these guys. Joining me now, ESPN’s Seth Walter, to look at the fast-changing field of O-line analytics.
Paul Michelman: Seth, thanks for joining the show.
Seth Walter: Yeah, happy to be here.
Paul Michelman: So let’s talk about the research. You’re tapping into a new source of data here. Is that right?
Seth Walter: Yeah. So the work that we’re doing here, the work that we’re talking about today, is based on the player-tracking data from the NFL, what the NFL calls its next-gen stats. There [are] chips in the shoulder pads of every player and [they’re] transmitting information about that player — their location, speed, orientation — multiple times per second, and that just gives us a plethora of information that we can work with [to] try to answer some questions that have been previously difficult to answer.
Paul Michelman: OK, so for our purposes, those questions have to do with offensive line play. So explain what you’re learning.
Seth Walter: So one of the first things that we did when we got access to this is… We have Brian Burke on our team, which gives us, I think, a pretty nice advantage. For those that don’t know, he’s one of the forefathers of football analytics. And so he was very excited to get his hands on this stuff. And one of the first things he wanted to do was to create something to try and measure offensive line play. Basically, if two players are standing near the line of scrimmage during the play, it’s a pass play, and they’re facing each other — well, it’s very likely that the offensive player is blocking the defensive player. And so once you have that information, we can start to break down pass blocking on an individual and team level. When a defender… when his chips in the shoulder pads, essentially, get past the offensive player and closer to the quarterback, we know that the defensive player has beaten his blocker.
What Brian created was something we call pass block win rate, something we unveiled in the middle of last season for the first time. I’m really excited about it. I’ve spent more time paying attention to offensive linemen over the past year than I ever have previously. And basically, what it does, is it says how often if you’re an offensive lineman, does the defender that you’re blocking beat you within two and a half seconds of the snap, which is roughly the average time to pass. And so we can measure that at basically an individual and team level. But the key here, that two and a half seconds, that’s a really important part of this question. Because when Brian first started looking at this... If you just looked at, well, how often does a certain player get beaten on a play? One thing that stood out to him was that in 2017, Joe Thomas, he noticed, ranked like 55th amongst offensive tackles in terms of how often he was getting beat on a play. And so he thought, well, that either means that Joe Thomas, who’s a future Hall of Famer, is either worse than we think by quite a lot, or we’re failing to capture something. And ultimately what was really going on is that Joe Thomas was blocking for DeShone Kizer. And DeShone Kizer holds the ball for a really long time. And so eventually, every offensive lineman is going to be beat. And so you have to kind of put them all on this level playing field of two and a half seconds or whatever you choose. But we chose two and a half seconds. And once you do that, Joe Thomas leapt up to be — I think he was third in pass block win rate that year.
Paul Michelman: So we’ve got a new source of data, we’ve got new analytics derived from that data, which are giving us a new way to judge the effectiveness of individual offensive linemen. Can we then spin that up one level and look at a unit’s effectiveness through this lens?
Seth Walter: Yeah, we can. So basically, if one player on the offensive line gets beaten within two and a half seconds, he’s lost. But also we consider the team to have lost its pass blocking for that play. Because if one player has a free rush to the quarterback, that obviously can greatly impact the QB on that pass play. So yes, we have a pass block win rate for every team in addition to every player.
Paul Michelman: Excellent. So now let’s take this up one more level. Can you correlate pass block win rate with the success or lack thereof of the team itself?
Seth Walter: Yes, and this is where things, I think, started to get really interesting. As the season went along last year, we noticed that just by the eye test, the good teams were kind of moving to the top, and the bad teams were moving to the bottom. And when the regular season ended, the Patriots, Rams, and Chiefs were all in the top four — that’s three of the four teams that ended up being played in the conference championship games. Eight of the top 12 teams in pass block win rate made the playoffs, and none of the bottom 12 teams in pass block win rate made the playoffs. So that was kind of interesting. And so what we did was then go back and look and say: OK, well let’s actually just look at a correlation between pass block win rate and expected points added per play, and pass block win rate and win percentage. What we found was that there was a correlation, there was definitely a correlation between some of these things. There was not that much of a difference, in fact, in correlation between pass block win rate and win percentage as there was for the entirety of defensive expected points added per play and win percentage.
However, that was looking at three years of data. We have pass block win rate back to 2016 —2016, 2017, 2018. If you look only at 2018, the correlation between pass block win rate and offensive success in winning was much higher. Now, immediately, I would think: Well, there’s no real reason why [we should] cut it off and just look at one season’s worth of data. Right? When you have three, [it] doesn’t make sense to just chop your sample. I think the really sort of interesting thing here is that there was a change in the data between 2017 and 2018. Zebra, the company that collects the data, made improvements to the way it is displaying orientation. And so we think that it’s possible, we don’t know, but it’s possible that pass block win rate is more accurate in 2018 than it was in 2016, 2017 — which has us wondering, you know, is this even more impactful than the three-year sample would suggest?
Paul Michelman: So given that the NFL is a pass-focused league, this does feel like a case where something we thought we knew to be true — the better the pass blocking, the more successful the team — gains even more evidence. But was there anything that surprised you about the data?
Seth Walter: Yeah, so we have the flip side of this coin, too. We have what we call pass rush win rate, which is basically the exact opposite, the inverse of the statistic. What if we look at it from the defense’s perspective? We can again do individuals and see how often they beat their blocker within two and a half seconds. It will surprise no one that Aaron Donald crushed the rest of the league in that metric. And then we can do the same thing for teams. I think what was interesting to me is that when we looked at the team-level data for defenses, the correlation between defensive success and pass rush win rate was not nearly as strong as it was for the inverse, for pass blocking and offense. And the same with pass rushing and winning. And so I think what I started to think about is maybe... that pass blocking is really the larger part of the equation when trying to figure out the pass blocking versus pass rushing interaction. And so that to me is pretty interesting, considering I think that’s not necessarily how we think about the game. You know, pass rushers, as far as defensive players go, can be stars. And we very rarely talk about offensive linemen as stars.
Paul Michelman: Is there any risk of confirmation bias here? So I’m thinking about how important Joe Thomas stands in the way you developed the analytic, based on the data. So you had something — Joe Thomas didn’t kind of fit that model. So you tweaked it so he did, which might suggest that [in] this analysis, environmental factors are playing a role here, that preconceived notions might be playing a role.
Seth Walter: I think there is a danger of confirmation bias, though I don’t know about in that instance. I should clarify I’m not entirely sure if Brian just used that as the example, or if he first looked at Joe Thomas and then tweaked the model. Either way, I think the part that has me wondering if we’re allowing a bias or some sort of effect to mislead us — I think it might be we... I’ve said to this point, carefully, that these things are correlated, but I haven’t quite said that pass blocking causes offensive success. And I’m wondering if it’s possible [this is] suffering from the same sort of thing that rushing attempts suffer from in that if you are winning a game, you are more likely to therefore have better pass blocking success — as opposed to if you’re losing a game, and the opponent knows that you need to pass, they can sort of blitz you without worrying that you’re going to run. Does that throw a wrench in the numbers? Of course, the flip side of that would also be true, that you would expect that the pass rushing then would correlate stronger with winning teams because of what I just said. And we don’t see that as much. And so I do think that there are questions here, no doubt about it. Are there things that are not necessarily clear? Yeah, I mean I think we’re still relatively early in the research process.
Paul Michelman: The data would suggest though, in your analysis of the data, that the correlation’s pretty strong. So how actionable are these findings for teams? You know, I’m thinking about draft. I’m thinking about free agency.
Seth Walter: I think quite actionable. But you know what? I think this is also one of the interesting questions that sort of comes out of this. You guys had Cade Massey on your podcast a few months ago, right? And his thesis was you can’t judge individual players in the NFL, because they’re all part of this larger ecosystem of an offense or a defense. And I think that that’s definitely a question here. Can we look at an offensive line and a single offensive lineman and say that he is more important than a single defensive lineman? I’m not sure that we have enough information yet to say that, because I think our working theory might be that what you really want to avoid on the offensive line is one weak link. That’s a big problem. And so, we see some examples of that like [with] the Dolphins. They didn’t have one weak link, they had three weak links, but they had two pretty good tackles in Laremy Tunsil and Ja’Wuan James. But because the interior of their offensive line was so weak, they were an awfully poor pass-block-win-rate team. Whereas, [on] the flip side, the Patriots had Shaq Mason, but no one else was really particularly elite on their offensive line. And yet, they were one of the best, if not the best, pass-blocking offensive lines in the league. So ... here’s my current theory, and this is what I would do if I were a GM right now: Basically, I would invest in the offensive line quite a bit, but I would focus that investment on mid-tier assets at depth. So not necessarily trying to pay for the best offensive lineman, but pay for many pretty good offensive linemen. And I would certainly think that it is very critical, I would say, in my opinion now (this is just how I feel) — I think it’s becoming more clear to me [that] the offensive line may be the second most important position group after quarterback. And so I would want to make sure I have a good offensive line. But whether that comes through elite players or just sort of average to good players is still up for debate.
Paul Michelman: Yeah, it’s interesting, right? Because when so many teams continue to pursue expensive free agents, there’s a cost to that. They’re going to have to underpay some people on the team. And if that’s on the offensive line, then that would run totally counter to what you’re seeing.
Seth Walter: Yeah, exactly... You may not always have the option when it comes to free agency. There may be only a few guys out there that are good enough. And so you do have to pay them a lot. So I do think that overall it is worth the investment to pay your offensive linemen or to invest in your offensive linemen. I just don’t know whether the elite players can contribute in an outsized way compared to an average player. I could be wrong about that. I mean, it certainly could be that if you have an elite tackle, that the guard next to him — his level of play — could be brought up, because you don’t have to help that tackle. I think that’s worth investigating. We just haven’t gotten there yet.
Paul Michelman: So on that note, what’s next in the research?
Seth Walter: Well, I’ve done a little more looking recently, in just the past few weeks, trying to look at the stability of these two things. Pass block win rate — I wanted to look at... the correlation between team pass block win rate in the first half and team pass block win rate in the second half, or team pass block win rate in the first half and offensive success in the second half. I mentioned this disparity between 2016, 2017, and 2018, previously. That was one area where there was a big disparity. There was a lot of correlation this year between first half pass blocking and second half offensive success. But that was not true in 2016 and 2017. Pass rush win rate on a week-to-week level wasn’t quite as stable when Brian looked at just a week-to-week correlation. But then I did a first half, second half look at pass rush win rate, and it was about as stable in that measure as pass blocking. So there is something there, it just doesn’t have the same impact that pass blocking does. I think the real question, the real thing that we want to get to is... what if we incorporate this into a predictive model? How much can this add to what we talk about with… You know, at ESPN we do a lot of predicting of who’s going to be good. What if we added this into an equation to try and figure out who will be good going forward?
Paul Michelman: Yeah, that’s really interesting. And you would assume that the technology as it gets better and better, and the data that emerges from it gets better and better, that the ability to predict based on that is going to increase. But it also raises a question of how relevant some of these kind of longer historical analyses are becoming when you know that the present day, whatever you’re getting right now, is of superior quality than what you got last year or two years before that.
Seth Walter: Yes. I think that’s an interesting part. I think that this sort of concept of: Well, the data is different in 2018, and how are you supposed to draw conclusions from this single year? How do you weigh one year over the next because of improvements in the data? It’s an interesting question. And the cool thing here is that especially in the NFL, where we got access to the data at basically the same time the teams did, is that if teams are building their own models along the lines of this, they’re also wrangling with that exact same question.
Paul Michelman: Great. Seth Walter, thank you very much.
Seth Walter: Thank you guys.
Paul Michelman: This has been Counterpoints, the sports analytics podcast from MIT Sloan Management Review. You can find us on Apple Podcasts, Google Play, Stitcher, Spotify, and wherever fine podcasts are streamed. And if you have an idea for a topic we should cover or a guest we should invite, please drop us a line at firstname.lastname@example.org.
Counterpoints is produced by Mary Dooe. Our theme music was composed by Matt Reed. Our coordinating producer is Mackenzie Wise. Our crack researcher is Jake Manashi, and our maven of marketing is Desiree Barry.