Jarvis Jones and Trade Value: How Football Analytics Can Hack the NFL Draft
This goes against a certain narrative, but the tale of the introduction of analytics into most sports is that of love at first sight.
Sure, there are head coaches who don't want anything to do with it. (Former Philadelphia 76ers and Detroit Pistons coach Doug Collins once famously explained that he'd rather "blow his brains out" than be an "analytic.") There are teams, like the Philadelphia Phillies and Los Angeles Lakers, that only pay it lip service. But for the most part, analytics gained widespread acceptance and became a tool in the toolkit pretty quickly.
Except, that is, in football.
.png?w=256)
.jpg?w=256 "Colts Jaguars Football")
Football is a bit more complicated than these other sports. There are more players on a team. There's more choreography to mess up. There are more moving parts, each of which can have a big impact on the play. Without the actual play calls and the knowledge of what each set of coaches is looking for from each player, breaking down the actual game of football is a case of educated guesswork.
There's also the fact that, unless you count former San Francisco 49ers head coach Bill Walsh as a man before his time, no "analytics team" has ever concocted the magic formula to win a Super Bowl. And, well, the NFL is such a copycat league that few teams are willing to risk being out on an island.
It's one thing to lose with the tried-and-true approach—dare to go against the conventional wisdom in the NFL and, the story goes, you will become a laughing stock.
One thing to remember about the concept of analytics is that analytics doesn't have to be pigeonholed. Was Walsh an analytics coach? He used data, and he critically thought about that data. Nobody was around to call him an analytics coach, so he got the term "genius" slapped on him instead.
There aren't "analytics teams" and "prehistoric teams." Creating a WAR system for the NFL or something like that isn't the only goal.
Analytics can be more broadly defined as "thinking through the available data before coming to a theory or conclusion," and it doesn't have to lead to a bulletproof hypothesis. NFL analytics can be as simple as noticing that a certain team is often out of position against screen passes and as complex as the number of calculations that go into Football Outsiders' DVOA formula, with plenty of degrees of separation in between.
So, that said, here are a few ways analytics can help NFL draft analysis.
Jarvis Jones, Michael Sam, and Athleticism
Using the publicly available football analytics, it would be hard to concoct a real case between say, A.J. Green and Julio Jones in the 2012 NFL draft. What we have at our disposal doesn't do a good job of differentiating between individual players—it does a great job of differentiating between large groups of players.
The best recent example is the case of Georgia outside linebacker Jarvis Jones, whom the Pittsburgh Steelers selected in the first round of the 2013 NFL draft.
The Steelers were thrilled to select Jones with the 17th overall pick, even cheering when his poor 40-yard dash time at the NFL combine made him more likely to be available, according to The New York Times' Andy Benoit:
"Colbert said hours after drafting Jones that he and his staff were happy when Jones ran a slow 4.9 in the 40 at his Pro Day. That’s because Colbert knows that there are N.F.L. teams that will drop a player down their draft boards for that.
Common sense suggests that how fast (or slow) a man runs 40 yards down a straight line while wearing spandex should have little bearing on how you feel about him as a football player. After all, what a 40-time can’t measure is how fast a player is when he’s wearing a five-pound helmet, 10-15 pounds of padding, starting his motion not from a runner’s crouch but an outside linebacker’s stance, changing directions in his movement all while being preoccupied with his assignment and with what his opponent is doing. Every team knows this. But for reasons no one will ever figure out, many teams can’t help letting times in the 40 sway their opinion about a player. The Steelers can.
"
And Benoit (now working for The MMQB) is right: There is plenty of evidence that a 40-yard dash time doesn't determine a player's worth. The problem for Jones wasn't that his 40-time was slow, it was that his everything was slow, short or poor.
The folks at MockDraftable have a large database of Indianapolis combine data—from the 40-yard dash to the three-cone drill—for every player since 1999. Here's a spider chart of how Jones' times fared:
Other than the broad jump, which Jones did at merely an average pace, he consistently hit near the rock bottom of every other combine drill. His highest measurement among the other drills ranked him in the 14th percentile among all outside linebackers.
Jones has played just 21 games and racked up three career sacks, looking slow the whole time. The NFL combine results revealed him to be an extreme outlier, which is one of the few things the publicly available analytics can hone in on, and the Steelers took the risk anyway.
Employing "analytics" in this case would've been as easy as putting all this information in a spreadsheet and seeing how few pass-rushers came into the league without any real athletic traits.
This, by the way, is the exact same reason that St. Louis Rams draftee and current free agent Michael Sam hasn't caught on with a team. It's not that he doesn't know the technique to play in the NFL. It's that he isn't athletic enough to be on par with NFL players, as this spider chart from MockDraftable shows:
NFL teams picked up on this, and that's why Sam's production in college didn't leave much of a dent on their radars.
It's not hard to overcome one or two bad traits. Some players will run faster in pads. Some players don't need to explode off the line if they can make it up with technique or win late in a down. But a player who struggles in every measurable physical attribute? They don't make much of an impact in the NFL.
The Brandon Marshall Baseline
If being traded for is a sign that a player is valued, then nobody is quite as valued as New York Jets wideout Brandon Marshall. The 2012 All-Pro has been dealt three separate times, bringing diminishing returns as he aged and became more expensive against the salary cap.
Analytics can't quite encapsulate the entirety of each Brandon Marshall trade, because opportunity cost as compared to the salary cap is a difficult beast to measure. Football Outsiders' Andrew Healey has tried, but attaching numbers to cap space is difficult because different teams have different free-agency goals.
What they can tell us is the average return of a draft pick in terms of AV over a five-year period. (AV, or approximate value, is a quick-and-dirty way of measuring player contributions over the course of a season. It isn't perfect, but it does let us measure players against each other a long way in the past.) And we can use that as a check-point to tell us if a trade makes sense. For instance, if we look at this Marshall trade:
| Brandon Marshall | ?? | Pick 142 | 3 |
| Pick 224 | 0.1 |
The Jets don't need to get much from Marshall to "win" the trade from the standpoint of AV. One could argue that the Jets could have used the cap space that Marshall takes up more efficiently, but given how poorly the Oakland Raiders, Cleveland Browns and Jacksonville Jaguars have used the huge allotment of cap space they've had over the past few seasons, Marshall adequately passes the smell test there.
And again, to remind you that this is just a checkpoint, a lot of other factors could impact how you feel about the trade. Maybe the Jets feel that a seventh-rounder is as good as a fifth-rounder this year. Maybe their non-Marshall option was to bring back Stephen Hill and roll over their cap space into the void.
We don't know everything that went into the trade from the outside. But looking at it in AV terms should, at least in theory, keep teams from getting ripped off.
Here's how Marshall's trade from the Miami Dolphins to the Chicago Bears looks:
| Brandon Marshall | ?? | 73rd pick | 7.3 |
| 82nd pick | 6.5 | ||
| Total | ?? | 13.8 | |
| 3 years of Brandon Marshall | 32 | Dion Sims | 3 |
| Fifth-round pick - Seventh-round pick swap | 2.9 | Michael Egnew | 0 |
| Will Davis | 0 | ||
| B.J. Cunningham | 0 | ||
| Total | 34.9 | 3 |
The two picks the Dolphins acquired should have combined for about 13.8 AV over their first five years in the league.
After all the wheeling and dealing, the Dolphins came up with two tight ends who have never cracked the top of the depth chart, a cornerback who has barely played and a wideout who didn't make their roster. Dion Sims has been worth three AV thus far in his career and if left to a starting job, could probably reach about 14 by the end of his fifth year.
All Marshall did was accumulate 32 AV all on his own in Chicago, before we even get into whatever the Bears win from their swap with the Jets. It's easy to dismiss the idea that the Bears won this trade simply because the Dolphins drafted poorly, but these figures are built on the average of all draft picks.
If you think the Dolphins picked poorly, then the entire NFL "drafts poorly." The Dolphins came out with an average result of those picks.
Again, it's important to factor in the idea of cap space, because the opportunity cost of that cap space does matter. But what this quick-and-dirty method can give us is a baseline to determine how fair or unfair trades are on their face.
Then we can get in to all the adjustments that won't find their way into the box score.
The Evolution of SackSEER
Even in smarter circles, regression analysis still faces a bit of a stigma. The idea of taking out a bunch of numbers to find the most "sticky" ones is a field that is usually rife with the catchphrase "correlation does not equal causation."
One of the best systems out there is SackSEER, a Football Outsiders project via Nathan Forster that attempts to put a number on how many sacks a player should have in his first five years in the NFL based on college production. Well, it was originally based on college production. Then, an interesting evolution happened.
Jason Pierre-Paul was the Moby Dick of SackSEER. He received one of the worst projections of any player coming out in 2010, which sent Forster back to the drawing board.
The revised statistic showed a lot of the same influence you saw further up in the article. An updated selection of the traits that best identified edge-rusher talent included a stat—passes defensed—that showed off college athleticism. It also would grow to include the three-cone drill into its model heavily.
Essentially, SackSEER was proven to be smarter as it included the same sort of athleticism "knockout factors" that Jarvis Jones and Michael Sam suffered from. (Jones, for the record, received a projection of just 23.7 sacks over five seasons, largely because of his awful athleticism. Sam received a projection of zero sacks.)
It's easy to initially be skeptical of regression-based systems. An initial skepticism in any model is a healthy trait to have, especially if you are the creator of the model.
| 2010 | 3.8 sacks in five seasons |
| 2012 | 19.7 sacks in five seasons |
But skepticism should help evolve models, and in this case, SackSEER is a great success story for skepticism. It went from a purely stat-based dart toss to something that incorporates a player's success and his traits.
It's not a guidebook to follow—and it never will be. But the model has improved to the point where an enormous outlier should be evidence that teams should second-guess a grade on a player rather than second-guess the model itself.
There's a saying I learned in creative writing classes that talks about how we only really know the "tip of the iceberg" at any one time in a great story. Gradually in a story, the water recedes, and we begin to learn more about the characters.
This is exactly where we're at with football analytics.
Not every team is working on them, but the teams that are have kept the vast majority of them under the surface of the ocean. It's to the point where the only things you'll see actual football executives talk about on Sloan Analytics Conference panels are theories they've already discredited.
So what we have is the surface. And the surface is gritty. It doesn't create beautiful and flawless analytics, because almost nothing in football does.
But it does create checkpoints that should cause concern to front offices when a player fails them. It should concern general managers that Shane Ray ran a slow 7.71 three-cone drill at his Pro Day. It should concern Philadelphia Eagles fans that Chip Kelly is willing to wager a second-round pick (with an average AV of 9.4) on Sam Bradford (who has zero seasons with an AV above 9).
That doesn't necessarily spell doom and gloom for the team that drafts Ray or the Eagles. But it does send up a red flag that both moves could be catastrophic failures.
If analytics is still mostly the iceberg under the water, then what we can see on the surface is simply the few warning flares that stand out.
All DYAR and DVOA numbers cited are courtesy of Football Outsiders. Learn more about DVOA here.
Rivers McCown is an NFL Analyst for Bleacher Report and the co-host of the Three-Cone Drill podcast. His work has also appeared on Football Outsiders and ESPN.com. Follow him on Twitter at @riversmccown.
.png?w=480)
.jpg?w=480 "Colts Jaguars Football")
