Welcome to the first installment of my rushing study on DLF. You can find my introduction, explanation, and numerical breakdown of the series here. In this two-parter, we’ll be looking at running backs that the community generally deem to be talented that run behind poor offensive lines. Todd Gurley and Melvin Gordon will serve as the central focuses and lead into a greater conversation about the constraints that a poor blocking unit can place on any back.
Todd Gurley, RB LAR
Gurley has been the poster boy for the good player-bad line confusion. In a destitute offensive situation exacerbated by poor coaching, Gurley drowned in 2016–he finished the season running at an impressively-terrible clip of 3.2 yards per carry. Despite a strikingly bad season of production, his dynasty value has hardly taken a hit. The implied reasoning behind that (lack of a) trend is that the heinous production was caused by his surroundings.
However, the data I’ve collected strongly disagrees with this sentiment. Why? Simply, he didn’t create at all:
[am4show have=’g1;’ guest_error=’sub_message’ user_error=’sub_message’ ]
The plot above shows each back’s success rate (ability to hit a hole) plotted against his Generated Yards Per Hole Hit (creativity and explosiveness). Among each back sampled, only Lamar Miller could approach (let alone, exceed) Gurley’s lack of creativity and explosiveness. Over his first 14 of 16 games, Gurley could only generate 4.3 yards for each hole he hit. For reference, Ezekiel Elliott, Jordan Howard, and LeSean McCoy doubled that mark.
When we look at fully-sampled backs, Gurley distances himself from the pack even more:
Gurley’s abysmal Generated Yards Per Hole Hit mark was an entire standard deviation away from second-worst Devonta Freeman. For the non-mathematically-inclined, that’s really bad.
Why didn’t he generate more yards? Well, there are a few ways to answer this question. As I explained in a past article, creativity comes in many shapes and forms. Look at how many different running attributes go into the GY/Hit metric:
In a few words, you could say he was collectively poor at the traits toward the bottom of the diagram. My current set of observational statistics have few ways to quantify those specific traits, but I do have a statistic for tackle breaking, which can help a lot when explaining the poor creation.
My Broken Tackles Per Hole Hit (which, obviously, represents the number of tackles a back breaks for every time he’s found daylight) has a pretty strong relation with GY/Hole Hit; the R-squared coefficient between the two measures comes in at .58. As you can see, Gurley (at the very end) had a very difficult time breaking tackles:
It’s a tall task to create yardage when you can’t break tackles… and Gurley could hardly shake anyone.
Gurley’s other open-field issue was one simply of athleticism. His speed and acceleration appeared rather ordinary across most of his carries, preventing him from adding explosiveness to his touches. It’s hard to keep GY/Hit numbers up without breaking tackles, but it becomes impossible to do so if there are no big plays to buoy a greater collection of small gains. Of his 245 carries that I have play-by-play data of, Gurley generated ten or more yards… four times. None of those contributions reached past the teens. That won’t cut it. His just-fine success rate (85.9%) essentially made him good enough to take what his offensive line gave him, but that was it.
Of course, that line was far from blameless itself. Let’s look at how that unit gra-oh my goodness that’s bad:
As you can see, the Rams offensive line was one of the worst in terms of OL-Generated Yards Per Carry (a baseline of yards for the back to carry) and boasted the very worst Hole Rate (how often the entire unit creates opportunities). They also graded out dead-last in my OL Composite statistic, which simply combines those two measures with equal weight. This unit was terrible.
Was there anything good about last season, then? Well, it could’ve been a bit worse. He was alright at reaching holes, posting a Success Rate that wasn’t very far below average. If he were legitimately bad, we could’ve seen the rare beautiful disaster that is 278 carries going for under three yards a pop.
Otherwise, the only good spin you could put on the season is that it’s probably impossible for those carries to be that unproductive again. The passing game can hardly get worse, and that in itself could mean more rushing production with defenses unable to sell out to stop the run on as many plays. Furthermore, I’d expect further improvement from a coaching change, as Sean McVay should at least instill a modern offense in Los Angeles. Most importantly, though, the arrows are pointing up for both rushing units.
The offensive line sees plenty of continuity, essential for blocking success, heading into 2017 as the only leaving contributors are Greg Robinson (not one of the better second overall picks in NFL history) and Tim Barnes (a former UDFA who inked a one year, $800,000 contract with San Francisco). Meanwhile, the Rams have brought in a small transfusion of talent that can be used to replace those two in Andrew Whitworth (solid, but old) and John Sullivan (who I know nothing about). Whitworth’s best days are behind him, but he should provide much steadier play than Robinson, at the very least. Sullivan is replacing a center who currently sits third on the 49ers’ depth chart. The bar for improvement is incredibly low.
From what I researched on Gurley in 2016, he didn’t look to be hobbled by injury very much, so health-based progression shouldn’t be expected. However, he’s a former first-round pick who rushed for 4.8 yards per carry as a rookie entering just his third season; there’s plenty of room to positively regress to the mean and genuinely improve even if he doesn’t tap all of his potential.
Of course, rushing production after progression from both sides could still be quite bad. The flipside to “lots of room for improvement” is that there’s a lot of room to make up. By my estimates, with moderate progression from the offensive line, Gurley would have to approach average (a tougher task than it sounds when considering where he ranked in 2016 and how good running backs are these days) at both Success Rate and Generated Yards Per Hole Hit in order to just get to four flat yards per carry.
In summary, Todd Gurley did run well in 2016 and when combining that performance with a terrible set of blockers, the running game suffocated. There are positive and negative ways to look at the future. On the bright side, there’s loads of room to improve if solely through regression to the mean, let alone through on-paper improvement to the offensive line. The bleaker outlook is that blocking will still probably be subpar, meaning the onus falls on Gurley to provide effective rushing production. Given his previous campaign, I’m not as optimistic about this happening as most are.
Check back in tomorrow for part two.
[/am4show]
Stephen Gill
Latest posts by Stephen Gill (see all)
- How Well Do We React to New Rookie Information? - October 2, 2019
- Toward a Comprehensive Model for Rookie Receivers - September 24, 2019
- Rookie ADP, Part Two: Who are you Lower on? - May 28, 2019
Ari Daum
August 16, 2017 at 6:56 am
We would love to see this data for 2nd tier backs like Montgomery, Powell, etc!
Stephen Gill
August 16, 2017 at 2:16 pm
Good to know! I’m hoping I can get around to those relevant guys too.
Ben Johnides
August 16, 2017 at 10:35 pm
Stephen, well-written article. I really enjoyed the data driven approach to examining Todd Gurley and the Rams OL! Despite Gurley not producing well, I like to think Jeff Fisher is to blame for his woeful production but that’s the low hanging fruit; can’t quite observe, track, and measure that data.
I read through your method description that you linked and kudos for the transparency and well thought out process. Can tell you put a lot of time into this project. A couple of questions that came up after the read, and maybe I missed something: (i) how is the data for a play handled if the OL doesn’t create a hole (ii) is it possible to covary for (a) defensive package and (b) strength of defense (probably not enough power to do this, but seems like most studies on nfl data suffer from low power) (iii) have you considered using an aggregate “back-up RB” as opposed to the specific back-up for each player? I think the comparison can get muddy at multiple points here considering whether the back-up is the actual back-up or change of pace RB and whether the back-up is running against the same defensive unit, say, if the game is already out of hand (I know these games quite well as a long-time Lions fan). Just some thoughts I had.
Great work; looking forward to reading your future articles!
Stephen Gill
August 18, 2017 at 7:06 pm
Thanks for the time reading and well thought-out response!
1- Part 1 (linked in the beginning) shows the formula for each stat, but I’m betting you’re wondering about RB-GY and its family of stats, which were admittedly ambiguously-worded in that write up, so I’ll try to clear that up. RB-GY/Hole Hit (the creativity stat used above) *does* add the GY from non-hit hole plays to its total as things stand, which is suboptimal.
The logical way to go about it would be to *exclude* the GY from non-hit hole plays, but because of how I began recording data originally, I can’t separate those plays from the totals (I started off essentially by tallying totals for games then entering them, whereas I now enter play-by-play stats into my spreadsheet, which means I can now separate the stats). That way, we’d have a truer open-field sort of creativity measure, which is more fitting to how I describe the stat. I’d also have a new stat for the GY for plays where there *wasn’t* a hole hit, which would effectively measure how a player creates with nothing offered.
As things stand now, this new stat and what I’ve basically described RB-GY/Hit to be are lumped into one. There’s some redundancy between the two (check the “What Matters?” flow chart from Part 2), so it’s not a huge problem with them combined, but as I’m moving on, I’ll be able to separate the two more and more. Really glad you brought this up.
2- Regarding strength of defense first, I’ve taken some lengths to study that already, and the results are actually fascinating. I’ve started off by comparing the variance both between each player and between an individual player’s games against Rushing DVOA (my defensive metric of choice), and plan to write about these results in the future. In short, some players are actually more influenced by quality of defense than others! Some of this I’m fairly confident is due to small sample size (just 14 games apiece), but the differences are fairly big which does seem to indicate some difference. The other finding, which I’m pretty sure about, is that OL play is influenced much more by defensive quality than RB play. So yeah, definitely keep an eye out for my writing (and newly-developed stats, I’m sure) in the future about this.
As for defensive package, it’s the chink in the armor. The dirty little secret of my data right now is that the Cowboys OL grades out between average and below-average, which is crazy and must be due to box stacking (especially when the Falcons OL grades *much* higher than everyone else). I began charting both box defenders and blocker/box defender advantage for each rush when I initially converted to play-by-play, but honestly, it added a ton of time. I could see what the early results of the box stats were, but I’m hesitant to do so with such limited sample size. I’m hoping to add that consideration once I get a bunch of time to chart again. Basically, it’s an admitted blind spot.
3- I’m not *quite* sure what this is asking, but I’m guessing that you mean when I look at generating yards and success rate, where the RB-GY are above the “replacement-level back” and success rate is partially determined by a “replacement-level back’s” ability. Both of these stats are based off hypothetical players and thus the standard isn’t higher or lower for any back that I chart, regardless of team. Nothing to this point has involved any team’s specific backup, so if I’m interpreting the question right, I think you might have glossed over one segment of this monstrosity of a series and misunderstood it. If I interpreted it incorrectly, please let me know!
Well that was a long comment! Thanks Ben, let me know if there’s anything else you’re wondering — I can get back to you quicker too if you e-mail or tweet me.
Ben Johnides
August 19, 2017 at 7:11 pm
Thank you for the thorough and thoughtful response! You answered all of my questions, quite well. And you were correct about point 3; I misunderstood what was going on there. Keep up the great work. Gonna dig into part two soon!