Thursday, July 14, 2011

Let's argue about running backs' statistics

At some point a few months ago, I stumbled across a K.C. Joyner article on ESPN.com; whatever it was seemed pretty interesting at the time. I'd never read his stuff before, and I was super impressed that somebody on a mainstream media website actually (a) understood advanced football stats and (b) used them to assemble some coherent and thought-provoking arguments. It seemed like the kind of content I'd be more likely to find at Smart Football than ESPN, and this made me happy.

Then he wrote about Denard Robinson. His conclusion (in fewer words): Denard will be just fine and dandy in a pro-style offense, because his yards per attempt on downfield passes (10-plus yards and 20-plus yards) was even better than Ryan Mallett's last year, and therefore he's basically an NFL quarterback.

This is called logic fail. Stats can be extremely useful; they can also be extremely misleading when used by people who don't understand what they mean or fail to include things like, you know, context.

I love me some Denard Robinson, but saying "he averaged lots of yards per throw so he's obviously a good downfield passer" is just plain lazy and wrong. This was essentially Michigan's downfield passing game last year:


And now to the other side:


Summary: Safeties sold out like mofos because they were terrified of letting Denard run past them at roughly the speed of sound, leaving Roy Roundtree to leisurely stroll down the hashmarks and enjoy some crumpets on his way into the endzone.

Those were good times (kind of), but they tell us pretty much nothing about Denard's ability to complete passes of any significant distance when defensive backs are actually, like, in the general vicinity of the receivers (for the record, he's not terrible at it, but he's not exactly Ryan Mallett).

At first I thought maybe that article was just an aberration -- it's pretty hard to find Michigan games on TV, after all -- but after reading a bunch more of his musings since that one in April, I feel pretty comfortable with the following observations about K.C. Joyner:
  1. He likes advanced stats and recognizes that they have value (yay).
  2. He doesn't understand how to apply them; sometimes (probably by luck) he's full of win and sometimes he just comes across like an idiot.
Which brings me to his most recent attempt at brilliance:
Knile Davis is the SEC's top RB: A look at the 'bell cow' rankings for SEC running backs shows a surprising leader.
Before I destroy K.C. Joyner, let me first say this: Knile Davis is a very good running back. He's probably one of the best in the SEC, and he was definitely overlooked last year because of the 14-foot-tall guy next to him in the Arkansas backfield who could throw lasers through brick walls and defensive players and whatnot.

THE DESTROY PART

Running backs are hard to judge with numbers. Emmitt Smith rushing for 40 billion career yards does NOT equal Emmitt Smith being the best running back in NFL history. Yards per carry, broken tackles, runs of 20-plus yards, etc. -- those are all swell and interesting, but they won't tell you anything about talent or competition or number of defenders in the box (well, they might do so indirectly, but you know what I mean).

These are the five numbers Joyner throws into a cauldron to create his formula:
  • Percentage of rush attempts with 10-plus-yard gains
  • Percentage of rush attempts with 20-plus-yard gains
  • Yards per attempt against teams with top-30 rush defenses
  • Yards per attempt against everyone else
  • Yards per attempt on plays with "good blocking" (obviously a subjective but probably more useful measure) against top-30 rush defenses
He ranks the six candidates in each category (six points for first place and one point for last), then adds up all the points and says "biggest number wins."

The result:
  1. Knile Davis (24)
  2. Michael Dyer (21)
  3. Brandon Bolden (20)
  4. Tauren Poole (19)
  5. Marcus Lattimore (12)
  6. Vick Ballard (9)
This is lame. Not that there aren't any interesting numbers in there -- the "good blocking" category at least attempts to strip out a massive variable -- but the formula is basically set up so whoever averages the most yards per carry and racks up the highest percentage of long runs (which can, in turn, severely skew yards per carry) is "the best." It's unfair because these numbers favor speed guys with fewer carries over guys who get the ball 30 times a game and have a lower yards-per-carry average because of the respect for their ability.

Wanna know who led all of college football in yards per carry last season? Hawaii's Alex Green, who went for 8.21 a pop and finished with 1,199 yards. Even if you were to ding him (maybe knock off two yards per carry) for playing against basically zero defenses with a pulse, he'd still have a ridiculous per-attempt average and a crazy percentage of long runs, way better than guys like Daniel Thomas and Jordan Todman (both just over five yards per attempt) last year. But nobody's arguing that Green is better than those guys, and for good reason.

One of the biggest things Davis had going for him last year (and will again this year): He's a complementary piece in a relatively pass-happy offense. I don't have access to full formation data, but I'm willing to bet that Arkansas spent at least 80% of its snaps last year in a three- or four-wide set, and those wideouts weren't decoys. Bobby Petrino wants to throw, and he wants to throw downfield. Having gaping holes in nickel and dime defenses is a nice side benefit.

For reference, think all the way back to Louisville's 2006 juggernaut -- pretty much the pinnacle of Petrino's offensive genius until last year -- when a thoroughly mediocre set of running backs (Kolby Smith, George Stripling and Anthony Allen) rode Brian Brohm's coattails to a seemingly impressive 5.3-yards-per-carry average that actually said a lot more about the offense than it did about the gentlemen running the ball.

There's a big difference between being the guy next to the quarterback and being The Guy.

THE PART WHERE I ARGUE ABOUT NUMBERS

Here's my biggest problem with all the numbers cited above: They in no way account for the way the defense is playing -- not the quality of defense, but the focus. I'm not saying there's a better number to do that; outside of tracking every snap and counting men in the box (or something similar), I'm not sure there is. But that's why we watch. I don't need to look at yards-per-passing-attempt numbers to tell me how good Andrew Luck is ... and if I did, I'd come out with a pretty inaccurate assessment since he finished 10th in the country last year (behind guys like Ricky Dobbs and Greg McElroy).

Some numbers don't pass the eyeball test, and it's usually because they're missing either data, context or both. This is either way above K.C. Joyner's head or irrelevant to him because it doesn't fit whatever argument he wants to make on a given day -- I'm guessing it's a healthy mix of the two.

THE PART WHERE I LOVE MARCUS LATTIMORE

I watched Marcus Lattimore last year. Marcus Lattimore could have started in the NFL last year -- yeah, as a freshman. I haven't said that since I watched Adrian Peterson.

He had 182 yards on 37 carries against Georgia, 184 yards on 27 carries against Tennessee and 212 yards on 40 carries (!!!) against Florida. He was the offense in most of South Carolina's games last season (at least the competitive ones) -- the options were (a) handoff to Lattimore or (b) Hilarious Stephen Garcia Interception TM.

When you're getting the ball on literally 60 percent of your team's snaps, your yards-per-carry average is not going to be super awesome -- those two things are pretty much mutually exclusive at any meaningful level of competition as defenses realize that you are kind of important to stop. That does not mean you aren't super awesome (and by "you," I obviously mean "Marcus Lattimore"). Lattimore gets the ball because he can grind out four or five yards on almost every carry against even the best defenses and even when everybody knows he's getting the ball, which is a massively important factor that's completely ignored in the statistics.

Seriously, just watch:


Don't try to convince me that Marcus Lattimore is the fifth-best running back in the SEC and barely half as good as Brandon Bolden.

Quick note: Football Outsiders put out some sort of FEI-based rankings last year for running backs, and I'm irrationally curious to see where guys like Lattimore come out this year. Those numbers still don't include any sort of adjustment for blocking help, defensive alignment or quarterback incompetence, but their value is that they directly compare each player's production against the expected production (on a per-play basis) and include a consistency rating (this is basically standard deviation, meaning a guy who alternates zero- and 10-yard runs will have a higher number -- and lower rating -- than a guy who gets five yards on every play).

THE SUMMARY PART

I probably would've let the whole thing go as a lacking-for-understanding-but-not-lacking-for-effort piece if Joyner had acknowledged that his numbers give us just a small sliver of the overall picture of how good these guys actually are. But this ...
As for Lattimore and Ballard, their showings illustrate how far these two have to go to reach the consistent performance level of the other top SEC backs.
... was where I drew the line*.

If he truly believes he just came up with the golden formula for running backs and next plans to tell me that Alex Green should have won the Heisman last year, I would kindly advise him to start, you know, actually watching football.

Also, get off my lawn. I like my stats contextual, dammit.

*I'm not alone here. Seriously.

0 comments:

Post a Comment

Powered by Blogger.