Valuing the play of an offensive line is naturally difficult. Unlike other positions, an offensive lineman's performance on any given play is marked by the absence of things, such as sacks and stuffs. It's possible to cobble together a kluge, combining a variety of things like sacks or runs that make it past a couple yards. While this would probably give us some idea of which lines have played better than others, it's the kind of inelegant stat I like to avoid whenever possible. I like simpler, more graceful stats with meaningful units I can wrap my head around. WPA is a stat like that, its unit simply being wins. So when I was thinking about offensive line performance, I started by thinking about wins and win probability.
The irony of an offensive line is that its function is completely defensive. When you think about it, the offensive line's job is to protect the ball carrier or passer from attacking tacklers. Whether it's a run or pass play, the defense is really the side on the attack, and it's the offensive line's job to defend against their onslaught. Blocking, by its nature, is defensive.
At the core, the offensive line's job is to prevent the defensive line, and in most cases the linebackers, from doing their jobs. The lower the success of the defensive front seven, the higher the success of the offensive line. In fact, there's not much more to it than that. It doesn't matter whether a block is a quick chip or a Michael-Oher-from-Blindside-drive-block through the back of the end zone, as long as it prevents the defender from making the play. Realistically, an offensive lineman can do no better than prevent his counterpart from making a play.
To measure offensive line performance, we start by measuring defensive play-making, which is precisely the purpose of +WPA. +WPA (and +EPA) measures defender play-making by summing the WPA of all the plays in which a defender is credited for a defensive victory. In other words, only plays that are setbacks for the offense are counted toward a defender's +WPA. The half of all plays that are offensive victories should not be charged against the defender who made the tackle. Consider the backside DE who tracks down a RB from behind 10 yards down-field. The WPA was positive for the offense, but at the same time the DE should not be penalized for his effort. (This concept is explained in depth here.)
Adding up the play-making impact of a defender, in terms of either WPA or EPA, produces +WPA and +EPA. Summing up the impact of an offensive line's opposing front seven defenders can tell us how well that line has performed.
For example, consider the Chicago Bears offensive line. We can add up all the +WPA of their opposing front seven (and exclude pass defense plays for linebackers). All their opponents' sacks, stuffs, tackles for losses or short gains, tipped passes, and quarterback hits that result in incompletions--anything that results in a setback for the offense--comes to a total of +0.57 WPA per game. These are all plays the offensive line failed to prevent, and we can say the Bears offensive line has allowed -0.57 negative WPA (-WPA) per game.
That -0.57 is meaningful because it's measured in terms of the impact on winning, but it only measures half of the story. It only captures the half of plays that result in setbacks for the offense, so even the best teams will have a negative sum. Going a step further, however, we can calculate that the NFL average so far in 2010 is -0.38 -WPA for all 32 offensive lines. Chicago's offensive line, therefore, is responsible -0.19 WPA per game more than the average team. After all that theorizing and number crunching, I can still wrap my head around that result: The Bears offensive line has played bad enough to cost their team a 19% chance of winning each game.
On the other side of the ledger are the Colts. Their offensive line has a -0.19 -WPA, which is 0.19 better than league-average. We can estimate that the Colts line helps their team win to a tune of +0.19 WPA per game (with more than a little help from their QB).
The usual caveats apply. Quarterback and running back performances obviously have much to do with the impact of the defensive front seven. Still, this is large part of the picture, and further, it does more than any other stat to isolate and highlight the contribution of the offensive line, the part of a football team least recognized by statistics. And of course, stats don't tell the whole story, especially when it comes to individual performance.
Below are two tables, the first in terms of -WPA and the second in terms of -EPA. These are measures of an offensive line's ability to prevent opposing front-sevens from making plays--sacks, tackles for losses or short gains, forced fumbles, QB hits, and so on. Each column lists results in two columns. The first considers defensive lineman impact only. The second considers defensive lineman impacts and linebacker impacts on run plays and pass rushes. -EPA is less context-sensitive, and likely more predictive, than -WPA.
-WPA/G for Offensive Lines
Rank | Team | DL only | DL & LB |
1 | IND | -0.10 | -0.19 |
2 | HST | -0.14 | -0.23 |
3 | NE | -0.14 | -0.27 |
4 | BUF | -0.13 | -0.28 |
5 | SL | -0.20 | -0.28 |
6 | JAX | -0.18 | -0.29 |
7 | SEA | -0.16 | -0.30 |
8 | ARZ | -0.21 | -0.31 |
9 | NYG | -0.21 | -0.31 |
10 | CIN | -0.19 | -0.33 |
11 | TEN | -0.22 | -0.35 |
12 | DAL | -0.23 | -0.35 |
13 | PHI | -0.25 | -0.36 |
14 | DEN | -0.24 | -0.37 |
15 | NYJ | -0.20 | -0.37 |
16 | TB | -0.21 | -0.39 |
17 | SD | -0.24 | -0.40 |
18 | ATL | -0.21 | -0.41 |
19 | GB | -0.24 | -0.41 |
20 | OAK | -0.26 | -0.41 |
21 | DET | -0.32 | -0.41 |
22 | NO | -0.19 | -0.41 |
23 | PIT | -0.27 | -0.42 |
24 | SF | -0.29 | -0.42 |
25 | CLV | -0.26 | -0.42 |
26 | MIA | -0.21 | -0.45 |
27 | MIN | -0.28 | -0.46 |
28 | KC | -0.33 | -0.49 |
29 | BLT | -0.30 | -0.50 |
30 | CAR | -0.37 | -0.50 |
31 | WAS | -0.34 | -0.53 |
32 | CHI | -0.45 | -0.57 |
-EPA/G for Offensive Lines
Rank | Team | DL only | DL & LB |
1 | IND | -3.61 | -6.86 |
2 | NE | -3.46 | -7.65 |
3 | NYJ | -4.69 | -8.63 |
4 | HST | -6.06 | -9.25 |
5 | MIA | -4.88 | -9.91 |
6 | NO | -5.07 | -10.41 |
7 | DET | -8.42 | -10.45 |
8 | BUF | -4.76 | -10.50 |
9 | PHI | -7.70 | -10.84 |
10 | CIN | -6.94 | -11.00 |
11 | ARZ | -7.20 | -11.15 |
12 | GB | -7.06 | -11.18 |
13 | CLV | -6.54 | -11.28 |
14 | PIT | -7.94 | -12.11 |
15 | DAL | -8.21 | -12.14 |
16 | ATL | -6.44 | -12.31 |
17 | TEN | -8.58 | -12.73 |
18 | KC | -8.63 | -13.09 |
19 | BLT | -8.03 | -13.22 |
20 | MIN | -7.78 | -13.27 |
21 | SEA | -8.38 | -13.79 |
22 | CHI | -11.21 | -14.30 |
23 | TB | -8.36 | -14.53 |
24 | SF | -10.10 | -14.61 |
25 | JAX | -8.93 | -14.72 |
26 | DEN | -8.86 | -14.76 |
27 | NYG | -9.95 | -14.76 |
28 | CAR | -10.85 | -14.80 |
29 | SD | -9.11 | -15.13 |
30 | SL | -10.55 | -15.44 |
31 | OAK | -10.26 | -15.82 |
32 | WAS | -10.88 | -16.15 |
The idea here - seeing what isn't in front of you (measuring the negative space...the dog that doesn't bark, however you want to describe it) is brilliant. It's a tough notion to wrap your mind around but this is a hugely insightful metric.
Should secondary sacks and tackles for a loss also be counted?
but how much of this has to do with peyton manning and tom brady putting their offensive lines in the right position on every play, and then having a good sense for how to move around in the pocket and when to get rid of the ball, vs. jay cutler and jason campbell standing idle and patting the ball leading to a sack?
Wow, this looks useful.
Perhaps more context is needed: DL included alway, LB on running plays or sacks (?), and CBs only on sacks (?). Does that make sense? LBs usually aren't the O-Line's responsibility on a pass play.
Also, it might be good to convert this into something centered around 0.
"it does more than any other stat to isolate and highlight the contribution of the offensive line"
This stat clearly does more than all the conventional stats, but how does it rate against other advanced line stats? Do you have some evidence for your above statement?
@DSMok1
Brian did separate LBs in coverage, and DB sacks are rare. I think Brian's caveats should include scheme (5 linemen protection is very different than max protect), but that might fall into standard caveats.
There are some values that affect winning that appear to be left out as well. QB hits in the act of throwing almost always result in negative WPA, but aren't accounted for. Also, despite the final sentence of the post, if sacks factor into much of the WPA/EPA values, neither of these stats are going to be particularly predictive. Sacks are not a good predictor of sacks. QB hits + Sacks + QB hurries is considerably better.
Overall, it's a nice additional stat.
Very cool post, Brian.
I've been saying for years that Indy's O-line is a huge part of what makes our offense great. They're not just good, they're the best. Thanks for the stats.
First off I like the thought process that went into this and how it was explained. Looking at the offensive line as a defensive mentality is something I hadn't thought of before.
I'm not sure if this is so much a measure of the offensive line but a teams run and pass blocking. On the majority of runs, there is a full back and/or tight ends blocking as well as the offensive line, and on pass plays there is frequently someone there helping as well.
I also think this is a step in the right direction but there are a lot of other variables too so this gives just a general idea. I'd include DB's on pass plays that are blitzing as well, because the OL/pass protection needs to pick that up.
I'd be curious to see this broken into pass plays and run plays, and see how much that correlates to teams with good QB's/good RB's.
And a last note, I agree with tgt, in that a combination of hits/sacks/hurries do a bit of a better job predicting sacks, and that some combination of that is probably a pretty good indicator of how good a line is in pass protection.
Regarding the question about how much a QB or RB affects these numbers--I don't know. Probably quite a bit. But now we have a tool that might help us measure how line play changes when the QB or RB changes.
To answer the question about what makes this stat better than other advanced OL stats, I would say that this approach:
-measures line play in a meaningful way in terms of net points generated or in terms of chances of winning games
-with a common frame of reference (WPA or EPA), it allows us to compare the importance of line play with the importance of other facets of the game
-we can also break the stat into 'pass rush' performance and 'running game' performance with a common frame.
-this also allows us to add in other effects, such as OL penalty WPA and EPA, and then compare their impacts
-other advanced stats I have seen are amalgamations of various things, such as sacks and running gains, which are weighted arbitrarily. (120% for a rush loss, 100% for no gain, 80% for 1-2 yard gains, or 100% for a sack, 50% for a QB hit, etc.)
This is great stuff Brian.
Firstly, I just wanted to check whether INTs were included in these stats. A good offensive line might cause more INTs but it would seem a tad unfair.
Secondly, it would be interesting to see how the O-line WPA for each team compares to the WPA for all the skill positions. Low correlations would mean that we successfully seperate the various components of the offence.
Not that this has anything to do with this post, but I'm hoping you might weigh in on what I thought was horribly bad play call in the Saints/Steelers game.
The Saints had 4th and 3 on the Steelers 13 with no timeouts and 12 or 13 seconds left in the first half. The Saints lined up for a field goal and then changed it up to show that they were going to run a play. The Steelers called a timeout, and after the timeout, the Saints kicked the field goal.
In the post-game press conference, Sean Peyton confirmed that the Saints were actually prepared to run a play and weren't just trying to get the Steelers to take a timeout.
I guess my question is this: would this fake field goal be a) a boneheaded call, or b) the most boneheaded call ever? I guess that Peyton might have had some kind of spectacular play called, but it would have to be pretty spectacular to overcome the horrible odds that were against it. Only a touchdown (from the 13 yard line) would have given them a much better expectation than they had before the play, and most other results would have given them a far worse expectation.
The announcers, of course, complimented Peyton on his Belichekish aggressiveness. I'd like you to weigh in on this because someone needs to point out the difference between aggressive play calling and just plain bad play calling.
Chris-Interesting question. At the end of the half, it's a slightly easier analysis because you don't have to kick off and a FG is worth a full 3 points and a TD is worth a full 7.
FGs are good from the 13 91% of the time, so a FG attempt is worth 2.7 EP.
Therefore, to go for the TD, you'd need the EP to be at least as high as the FG attempt. So: 2.7 = 7x
x=0.39
If Payton thinks his team has a better than a 39% of getting the TD on the fake FG, then sure it would be a good call. There just aren't that many fake FGs that we can get a good estimate of a baseline probability of success.
From the 13, a normal play from scrimmage might have a better chance. But I doubt it. The baseline success rate for 4th and 13 inside the 20 yd line is only 20%. As good as Brees and the NO offense is, I doubt they'd have better than double the baseline success rate.
I'm quite surprised that Buffalo is a top offensive line. They are 4th by one stat, and 8th by the other.
The factor you forgot to mention besides quarterback and running back play influencing these numbers is the schedule. Obviously a teams offensive line numbers are going to be inflated if they're going up against a worse defense relative to other teams on average. are there any adjustments you can make? it might be simpler to just watch the games and observe how the line communicates, run block, handle speed/power rushers and note what kind of penalties they get.
If WAS O-line is 31/32 by this measure, that adds luster to RG3's performance so far.
also, if that's the case how did they end up as the top rushing team in the league? Based on their woeful stats here, and Alfred Morris so so stats here (middle of the league), where is this actual on field performance coming from?
published on 11/02/2010
I'd like to see another variable, OL stats with the play left the same vs. after a QB audible or a kill play. Just curious.
Thanks for the insightful look at OL.
As a former offensive lineman my teams measured what was called "Pancake Blocks" where you knock over the opposing defender without the use of a chop block (diving at the knees which is much easier to knock over an opponent vs the strength and leverage applied to knocking your opponent over normally) as a measurement of Offensive line run-block capabilities. I don't believe that the NFL keeps track of this, but it would be a cool metric to see and wouldn't be a part of the measuring of a negative. But overall very nice analysis.