Rise and Fall of a Pass Rush

By this point it’s no longer a secret that the pass rush is the most important part of a defense. Any casual football fan has seen this become clear over the past ten years, as offenses have become more oriented towards the passing game and defenses have been forced to adapt. Despite the success of the Legion of Boom, a pass rush remains the easiest way to slow down a passing attack, and to build a championship caliber team in the NFL a top notch pass rusher is almost a prerequisite.

So far this season has given us plenty of examples of how much a defensive front can affect a game. The two most disappointing teams in the league are the Seahawks and the Colts, both of whom are 0-2 due to their inability to stop the opposing defensive lines. I touched on these teams last week, and I don’t really have anything new to say. They had the misfortune of playing teams with loaded defensive fronts, and they’ll have more success as they face easier competition. The offensive lines will remain a serious concern that could hold them back once the playoffs arrive, but I’m not going to write any seasons off just yet.

We’ve also seen examples on the other end of the spectrum. The Bills demolished the Colts and held reasonably tight against the Patriots despite not being good at anything outside their front four. Carolina’s front continues to beat up on bad teams, and Denver has survived offensive woes on the strength of an absolutely terrifying pass rush that will keep them in the hunt all year no matter what ends up happening with Peyton Manning.

The winless teams and the undefeated teams make excellent demonstrations, but far more interesting are the teams who have split the results over the first two weeks. Specifically, three 1-1 teams have reached this record by virtue of inconsistent defensive fronts, carrying them to the top one week while dooming them to failure the other. Two weeks into the season we still don’t know what we will see from the defensive fronts of the 49ers, the Steelers, and the Vikings, but whichever way it goes will play a major role in their fates over the rest of the season.

These three teams have actually played a pair of games between them so far this year. The 49ers’ victory came over the Vikings in Week One, a dominating 20-3 victory on Monday night football that was followed by a 43-18 beatdown at the hands of the Steelers in Week Two. Pittsburgh was coming off an opening night defeat of the Patriots, while Minnesota followed up their loss to San Francisco with a 26-16 victory over the Lions.

For none of these three teams was it really clear what we were supposed to expect from their defenses coming into the season. We are all aware of the turmoil San Francisco went through this offseason, losing three of the four best players in their front seven. After years near the top of the league, many expected them to regress on defense, though there were glimmers of hope with the return of Navarro Bowman and young players like Aaron Lynch and Arik Armstead.

We had a better idea what to expect from Pittsburgh and Minnesota, but there were still questions coming into the year. The Vikings had one of the best fronts in the league last season, and many expected them to take the next step this year. Pittsburgh was on the other end of the spectrum. They were absolutely brutal a year ago, but there were promising signs for their front seven. After years of Dick LeBeau’s zone blitzing scheme that relied on their defensive linemen to eat up blockers while the linebackers shot into the backfield, the Steelers decided to move on from the Hall of Famer, promoting linebacker coach Keith Butler to defensive coordinator. Though they kept their defensive scheme mostly the same, they made changes to better fit their personnel, asking their defensive line to be more aggressive shooting into the backfield. With budding star Cameron Heyward, underrated nosetackle Steve McLendon, and promising second year player Stephon Tuitt, there was hope that this change would allow them to pose a greater threat to quarterbacks this year.

After the first week of the season though, the faith in the Vikings and the Steelers seemed misplaced. Pittsburgh couldn’t touch Tom Brady on opening night, allowing him to chew their secondary to pieces at a pace even their high powered offense couldn’t keep up with. And on the following Monday, the Vikings defensive front was beaten down by San Francisco’s offensive line. They couldn’t contain Colin Kaepernick, and they could do nothing to stop Carlos Hyde on the ground. Even though it looked like the 49ers were playing most of the game without wide receivers, they moved the ball without much trouble on the Vikings.

On the other side of the ball, San Francisco’s front looked better than even the most optimistic 49ers fan could have imagined. Teddy Bridgewater was running for his life the entire game, and Adrian Peterson managed only 31 yards on ten carries. Minnesota’s wretched and injury plagued offensive line deserved its share of blame, but there was no way to deny that San Francisco’s defense had the look of a high quality pass rush.

Of course, as often happens, everything we thought we knew after Week One fell apart in Week Two. San Francisco didn’t manage to so much as hit Ben Roethlisberger one time, and the Steelers probably could have scored even more than the 43 points they put up if they’d kept their foot on the gas.

While San Francisco’s pass rush regressed, both Pittsburgh and Minnesota took major steps forward. Matthew Stafford looked absolutely miserable on Sunday, hit so often and so violently that he actually had trouble figuring out which of his wounds was dripping blood down his arm. Pittsburgh found similar success against the same 49ers line that had shut the Vikings down a week earlier, compiling five sacks on Kaepernick and doing a fine job containing Hyde.

So what changed between the two weeks for these teams? Was it normal variance, or was there something more schematic involved? And, more importantly, which of these teams are we more likely to see over the course of the remainder of the season?

To figure out why things swung so much between the first and the second week, we need to look at what things had changed. The first and most obvious thing to look at is the competition. If a team plays a top quality offensive line one week and faces garbage competition the next, it’s obvious that their pass rush will improve.

In the case of San Francisco, this looks like the easiest explanation. Minnesota’s offensive line was expected to be bad coming into the season, and that was before they lost John Sullivan and Phil Loadholt to injuries. They were awful in the first game, and they weren’t all that much better against Detroit, only putting up better protection numbers by virtue of more quick passes. On the other hand, Pittsburgh quietly has one of the best offensive lines in the league, and even without center Maurkice Pouncey they have done a fine job keeping Roethlisberger upright. New England and San Francisco combined for only two sacks of Roethlisberger, while putting up thirteen sacks between their games against Minnesota and Buffalo. The difference between the offensive lines of the Vikings and the Steelers are drastic, possibly even enough to explain the disappearance of San Francisco’s pass rush.

Of course, this early in the season it isn’t always that easy to evaluate a team’s competition. The Vikings demolished Detroit’s offensive line, a unit that held up fairly well Week One against San Diego’s lackadaisical defensive front. Pittsburgh didn’t do much against the Patriots to start the season, and New England followed that up by holding down Buffalo’s elite pass rush. But things get interesting when you consider that both the Vikings and the Steelers faced San Francisco. The 49ers offensive line has been one of the top units in the league for years, and even with some losses they looked very good against Minnesota. Pittsburgh tore them to pieces, raising questions about how good they are and how bad the Vikings might be. Competition is useful, but this early in the season it is difficult to be sure what to think of any teams. So instead we should look for schematic changes that might have occurred between games.

Pittsburgh didn’t make many changes, but there were major differences between the offensive styles of the two teams they played. New England is perfectly suited to take advantage of Pittsburgh’s defensive weakness, which is their secondary. The Steelers actually managed to get good push against New England’s offensive line, but it didn’t matter when Brady was able to get the ball out before the rush could get anywhere near him. Against the Steelers secondary there are always going to be open receivers, and few quarterbacks in NFL history have been as good as Brady at identifying these opportunities prior to the snap and getting the ball to them.

The 49ers passing attack is more representative of the rest of the league. The openings were there, but Kaepernick wasn’t able to find them, allowing Pittsburgh’s front time to show how good they are. And given the opportunity, their young players stepped up in every way we could have imagined. Bud Dupree looks explosive and has a sack in each of his first two games. Stephon Tuitt added a pair of sacks against the 49ers. Most impressive was Ryan Shazier, who flew all over the field both against the run and the pass. He shot behind the line so quickly and so frequently that I began to wonder if he was actually starting each play in San Francisco’s backfield. What Pittsburgh did against San Francisco is a best case scenario, but it is more likely to recur than their struggles against Brady in Week One.

The biggest change the Vikings saw was on their own side of the ball. After playing only 10 defensive snaps against San Francisco, rookie Eric Kendricks saw 53 against Detroit, the second most of any linebacker on the Vikings. A large part of this decision was schematic, a heavier use of their nickel package against a pass happy Lions team. But after the success their front had with him in the lineup, the Vikings need to consider making this a full time change. Chad Greenway has been a fine player for a long time with the Vikings, but right now he is a liability whenever he is on the field, and it isn’t a coincidence that the Vikings had one of their best defensive performances in years when he played a career low 14 defensive snaps.

It remains to be seen how Minnesota will shape their defensive front going forward, but I trust that Mike Zimmer knows what he does and does not have. Some parts of their performance will fall off (Brian Robison is a mediocre player and always has been, and he won’t be nearly as good as he was against the Lions going forward), but like the Steelers they have a lot of youth that will continue to improve on that side of the ball.

We know how important a pass rush is to a defense’s success, but we often overlook how volatile this can be. We’re used to seeing the performance of a quarterback swing from week to week, but a unit as a whole should be safer from such randomness. But this clearly isn’t always the case. Pass rushes rise and fall as wildly as any other part of a football team, and these swings can play as great a role in a team’s fortune as inconsistent quarterback play.

Point(s) After Touchdown

We are two weeks into the era of the long extra point, and it has already produced some interesting results. An eighth of the way through the season we have seen more missed extra points than we did all of last year (a fact that was mentioned on just about every broadcast last week). But the most interesting part of all of this was what the Steelers did last weekend. On each of their first two touchdowns they chose to go for two rather than kick the longer extra point, converting both times and opening up an early lead that carried them into a blowout victory. No other team has shown this aggression so far, but that could change as the season goes along.

The math of this decision is interesting, if murky. For years now there have been debates about the logic behind the PAT decision, but it seemed mostly settled that kicking was the smarter option. NFL extra points have been converted at a rate of .988 since 1994 (as far back as Pro Football Reference’s play data goes). Two point conversions have been successful at a rate of .475 over that same time, resulting in .95 expected points, fewer than a standard extra point.

When the change was announced to move the extra point kick back thirteen yards, the immediate question was how this math would change. The short answer is, it’s not clear. Since the line for a two point conversion hasn’t moved, the expected points remain the same at .95. But the kick is now 33 yards in length, a distance that has been converted at a rate of .896 since 1994. By these numbers it would appear that the prudent move would be to go for two. But field goal kicking has taken an incredible leap in efficiency in recent years, and if you limit Pro Football Reference’s data to the past five years (a sample more representative of current kickers), the efficiency of 33 yard field goals jumps up to .946, a difference in expected points that can be explained by simple statistical variance.

Expected points is a useful metric early in games, when the added utility of each additional point is roughly equal . But late in games things get more complicated. Naturally when a team scores while down eight, the value of a successful two point conversion is worth more than twice that of a successful extra point. This is a simple, obvious deficit, one every football fan understands. Things get more interesting when a team is down by a number of points that isn't quite so neat.

Pittsburgh’s success this past weekend got me thinking of another instance involving the Steelers, late in their opening night matchup against the Patriots. Down fourteen points, Pittsburgh was driving for a touchdown that would have pulled the game within a single score. Normally in this situation a team would kick the extra point to cut the lead to seven (as the Steelers did when they scored with two seconds left on the clock). But with the way the math had changed, I started wondering if they would consider going for two, attempting to cut the lead to six points so an ensuing touchdown would give them the win rather than send the game to overtime.

It didn’t end up mattering, but my mind kept running through this after the game ended. So I constructed a very rudimentary win expectation model to try to figure out what the best decision is. To make this model, I had to make a couple of assumptions. The first is that when a game goes to overtime, each team has an equal probability of winning. This is a reasonable assumption backed up by observational evidence. So for a game that ends up tied, I’m considering that outcome as half a win.

The second assumption is that the decision being made does not affect the probability of other events in the game. Going for two doesn’t make ensuing touchdowns any more or less likely, and it doesn’t change whether the opposing team scores. This may not be quite true. If a team has a one point lead late in the game, the opposing team may be more aggressive trying to score in the remaining time than if the game was tied, slightly decreasing the .5 odds I mentioned above. This is a potential source of error, but I don’t think it is greatly significant.

With these assumptions, other probabilities end up dropping out. Regardless of whether a team goes for two or kicks the extra point on their first touchdown, their chances of scoring a second touchdown are the same. So the only numbers we have to take into consideration are the win probabilities made under the assumption that two touchdowns are scored and the opposing team does not score. In order to get these values (which I will refer to as “Scaled Win Probability”), I constructed outcome trees for each possible decision.

Using this chart, it’s fairly simple to figure out the win expectations. Assuming a probability of success on a kick of X, and the probability of success on the two point conversion of Y, we end up with the following values.

Extra Point Scaled Win Probability   =  .5*X² + .5*(1-X)*Y

Two Point Scaled Win Probability    =  Y*X + .5*Y*(1-X) + .5*(1-Y)*Y

Rather than just using the slightly unreliable probabilities I mentioned above, I created a table to show the difference between these two numbers for values of X between .85 and 1 and values of Y between .3 and .6. I subtracted the extra point value from the two point value to get the Scaled Win Probability Added, a measure of the amount a team's win probability would increase by going for two. I colored the table for ease of analysis, green cells indicating a team should go for two while red cells show when a team should just kick the extra point.

	0.85	0.86	0.87	0.88	0.89	0.9	0.91	0.92	0.93	0.94	0.95	0.96	0.97	0.98	0.99	1
0.3	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0.1	-0.1	-0.1	-0.1	-0.1	-0.1	-0.1
0.31	0.01	0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0.1	-0.1	-0.1	-0.1	-0.1
0.32	0.02	0.01	0.01	0	-0	-0	-0	-0	-0	-0	-0	-0	-0.1	-0.1	-0.1	-0.1
0.33	0.03	0.02	0.02	0.01	0.01	0	-0	-0	-0	-0	-0	-0	-0	-0	-0.1	-0.1
0.34	0.04	0.03	0.03	0.02	0.02	0.01	0.01	0	-0	-0	-0	-0	-0	-0	-0	-0
0.35	0.05	0.04	0.04	0.03	0.03	0.02	0.02	0.01	0.01	0	-0	-0	-0	-0	-0	-0
0.36	0.06	0.06	0.05	0.04	0.04	0.03	0.03	0.02	0.02	0.01	0.01	0	-0	-0	-0	-0
0.37	0.07	0.06	0.06	0.05	0.05	0.04	0.04	0.03	0.03	0.02	0.02	0.01	0	-0	-0	-0
0.38	0.08	0.07	0.07	0.07	0.06	0.05	0.05	0.04	0.04	0.03	0.03	0.02	0.02	0.01	0	-0
0.39	0.09	0.08	0.08	0.07	0.07	0.06	0.06	0.05	0.05	0.04	0.04	0.03	0.03	0.02	0.02	0.01
0.4	0.1	0.09	0.09	0.08	0.08	0.07	0.07	0.06	0.06	0.05	0.05	0.04	0.04	0.03	0.03	0.02
0.41	0.11	0.1	0.1	0.09	0.09	0.08	0.08	0.07	0.07	0.06	0.06	0.05	0.05	0.04	0.04	0.03
0.42	0.12	0.11	0.11	0.1	0.1	0.09	0.09	0.09	0.08	0.07	0.07	0.06	0.06	0.05	0.05	0.04
0.43	0.13	0.12	0.12	0.11	0.11	0.1	0.1	0.09	0.09	0.08	0.08	0.07	0.07	0.06	0.06	0.05
0.44	0.14	0.13	0.13	0.12	0.12	0.11	0.11	0.1	0.1	0.1	0.09	0.08	0.08	0.07	0.07	0.06
0.45	0.15	0.14	0.14	0.13	0.13	0.12	0.12	0.11	0.11	0.1	0.1	0.09	0.09	0.08	0.08	0.07
0.46	0.15	0.15	0.15	0.14	0.14	0.13	0.13	0.12	0.12	0.11	0.11	0.11	0.1	0.09	0.09	0.08
0.47	0.16	0.16	0.16	0.15	0.15	0.14	0.14	0.13	0.13	0.12	0.12	0.11	0.11	0.1	0.1	0.09
0.48	0.17	0.17	0.16	0.16	0.16	0.15	0.15	0.14	0.14	0.13	0.13	0.12	0.12	0.12	0.11	0.1
0.49	0.18	0.18	0.17	0.17	0.17	0.16	0.16	0.15	0.15	0.14	0.14	0.13	0.13	0.12	0.12	0.11
0.5	0.19	0.19	0.18	0.18	0.17	0.17	0.17	0.16	0.16	0.15	0.15	0.14	0.14	0.13	0.13	0.13
0.51	0.2	0.19	0.19	0.19	0.18	0.18	0.18	0.17	0.17	0.16	0.16	0.15	0.15	0.14	0.14	0.13
0.52	0.21	0.2	0.2	0.2	0.19	0.19	0.18	0.18	0.18	0.17	0.17	0.16	0.16	0.15	0.15	0.14
0.53	0.21	0.21	0.21	0.2	0.2	0.2	0.19	0.19	0.19	0.18	0.18	0.17	0.17	0.16	0.16	0.15
0.54	0.22	0.22	0.22	0.21	0.21	0.21	0.2	0.2	0.19	0.19	0.19	0.18	0.18	0.17	0.17	0.16
0.55	0.23	0.23	0.22	0.22	0.22	0.21	0.21	0.21	0.2	0.2	0.2	0.19	0.19	0.18	0.18	0.17
0.56	0.24	0.24	0.23	0.23	0.23	0.22	0.22	0.22	0.21	0.21	0.2	0.2	0.2	0.19	0.19	0.18
0.57	0.25	0.24	0.24	0.24	0.23	0.23	0.23	0.22	0.22	0.22	0.21	0.21	0.21	0.2	0.2	0.19
0.58	0.25	0.25	0.25	0.25	0.24	0.24	0.24	0.23	0.23	0.23	0.22	0.22	0.21	0.21	0.21	0.2
0.59	0.26	0.26	0.26	0.25	0.25	0.25	0.24	0.24	0.24	0.23	0.23	0.23	0.22	0.22	0.22	0.21
0.6	0.27	0.27	0.26	0.26	0.26	0.26	0.25	0.25	0.25	0.24	0.24	0.24	0.23	0.23	0.22	0.22

Wow, that’s a lot of green. Yes, under most probability combinations, going for two gives a team a better chance of winning. If we use the probabilities I mentioned above for the new rules (.95 for the extra point, .47 for the two point conversion), we get a return of .12 Scaled Win Probability Added, well above the break even point.

So it now becomes very clear that the Steelers (and all teams trailing by 14 late in the game) should have gone for two. But the most interesting thing is, this isn't an effect of the rule change. Under the probabilities of the old rule (.99 for the extra point, .47 for the two point conversion), the Scaled Win Probability Added was still .1, significantly higher than the break even point. In fact, even if the extra point was a guarantee, a team would still have to have less than a 37% chance of converting on two to be better off just take the point.

This is just one example, the sort of situation that will come up a few times every season. But it is just one of many situations that is not nearly as clear cut as we have always perceived it. The math of the conversion may not have changed significantly, but the fact that it has changed at all gives us the opportunity to look at the strategy of football from a new and better perspective.