Welcome to Regression Alert, your weekly guide to using regression to predict the future with uncanny accuracy.
For those who are new to the feature, here's the deal: every week, I dive into the topic of regression to the mean. Sometimes I'll explain what it really is, why you hear so much about it, and how you can harness its power for yourself. Sometimes I'll give some practical examples of regression at work.
In weeks where I'm giving practical examples, I will select a metric to focus on. I'll rank all players in the league according to that metric, and separate the top players into Group A and the bottom players into Group B. I will verify that the players in Group A have outscored the players in Group B to that point in the season. And then I will predict that, by the magic of regression, Group B will outscore Group A going forward.
Crucially, I don't get to pick my samples, (other than choosing which metric to focus on). If the metric I'm focusing on is yards per target, and Antonio Brown is one of the high outliers in yards per target, then Antonio Brown goes into Group A and may the fantasy gods show mercy on my predictions. On a case-by-case basis, it's easy to find reasons why any given player is going to buck the trend and sustain production. So I constrain myself and remove my ability to rationalize on a case-by-case basis.
Most importantly, because predictions mean nothing without accountability, I track the results of my predictions over the course of the season and highlight when they prove correct and also when they prove incorrect. Here's a list of all my predictions from last year and how they fared. Here's a similar list from 2017.
An Easy Win to Start Us Off
If you read last week's column, you know that one of the keys to profiting off of regression to the mean is recognizing that everything regresses, but not everything regresses at the same rate. The more a statistic is dominated by luck, the more that statistic is going to swing wildly from one sample to the next.
Because I'm going to be making predictions and tracking their accuracy, I want to start the season off with my best prediction, the one I'm most confident in. And to make that prediction, I want to focus on the statistic that is more dominated by luck and random chance than any other statistic I know. I want to focus on yards per carry (ypc).
Yards per carry is one of the most beloved statistics for judging running backs. Jamaal Charles has never averaged below 5 yards per carry in a season where he's had at least 20 carries*, therefore Jamaal Charles is a star. Trent Richardson had 1300 yards from scrimmage and 12 touchdowns as a rookie, ranking as a top-10 fantasy back, but his 3.6 yard per carry was an early warning sign that he would eventually be regarded as a colossal bust.
*(Technically, Charles averaged 4.97 yards per carry in 2013, but what's a few hundredths of a yard among friends?)
I've written more about Trent Richardson before, back in 2014 when another young rookie had just had a high-volume, low ypc season that had everyone drawing parallels and claiming he was destined to disappoint. I wrote that, based on history, maybe we shouldn't be writing off this Le'Veon Bell fellow quite so quickly.
Indeed, the list of high-volume, low-ypc rookie running backs was basically Trent Richardson and a who's who of Hall of Famers or almost Hall of Famers. In addition to Richardson (3.56 ypc) and Bell (3.52 ypc), there's LaDainian Tomlinson (3.65), Ricky Williams (3.49), Walter Peyton (3.46 ypc), Emmitt Smith (3.89 ypc), Matt Forte (3.92 ypc), and Marshawn Lynch (3.98 ypc).
Even the guys on the high-volume, low-ypc list who didn't go on to be All-Pros typically had several quality fantasy years in them. Karim Abdul-Jabbar, Travis Henry, Errict Rhett, and Joe Cribbs all followed up their “inefficient” rookie season with a top-12 fantasy campaign as a sophomore, Sammie Smith improved across the board and finished as RB18, and Jahvid Best looked (and produced) like a star before injuries derailed his career.
Since I wrote that article in 2014, Melvin Gordon has also found himself on the “wrong” side of the ledger with an awful rookie ypc of 3.48. Fearing the shade of Trent Richardson, many owners sold low on the “inefficient” Gordon after a “disappointing” rookie season, only to see him rank 3rd in fantasy points (nearly 20% ahead of fourth place) from 2016-2018.
Indeed, other than Richardson himself, the only running back who had a high-volume, low “efficiency” rookie season and followed it up with a disappointing sophomore campaign was James Jackson, who also happens to be the only back in the sample to average below 3 yard per carry as a rookie, (2.84), and whose team thought so little of him that they drafted William Green in the first round to replace him.
What is going on here? Why is having a terrible rookie yard per carry average such a positive sign for a player's career? The truth is that a poor yard per carry average isn't a positive sign. It just isn't a negative one, either. I'm providing a list of high-workload rookies with low yards per carry, and the high-workload part is the real key.
Backs get a high workload because the coaching staff thinks they're good and wants to give them the ball. In the long run, backs who coaching staffs think are good and want to give the ball... tend to be pretty good. The low ypc, in the meantime, is just a meaningless fluke.
What Is Yards per Carry, Anyway?
To understand why yards per carry is a fluke, you have to understand something very important about it: it's not measuring how good a running back is. It's so thoroughly dominated by outlier runs that all it's really measuring is whether a back has had three long runs or merely two. 28 running backs beat the league's ypc average last year; 15 of them would have been below average if not for one or two long runs. Similarly, how many backs were just one broken tackle away from going from a below-average ypc to an above-average one?
To some extent, long runs are a product of player skill. But they're a product of a very specific skill— straight-line speed. Someone like Le'Veon Bell might excel at every other skill required of the position, but since he lacks high-end straight-line speed his ypc will always underestimate his value. Indeed, the longest touchdown run of Bell's career is just 38 yards.
To an even larger extent than skill, long runs are a product of luck. First and foremost, you can't run for 50 yards if your team is only 40 yards from the end zone. Additionally, you likely need some combination of good blocking and poor tackling to get into space in the first place so you can put that straight-line speed to good use. And insofar as long runs are dominated by luck, you'd expect them to vary wildly from one sample to the next.
What does this mean in practice? Statisticians have a concept called “face validity”. Most of the rest of us better know it as “the smell test”. Let's say I invent a statistic that I claim measures how good running backs are. The first thing I should do is look at a list of running backs under my new statistic and see if my statistic has face validity— see if it passes the smell test.
If I ranked the 75 running backs who have 500 carries over the last decade, and I told you my #1 back was Jamaal Charles and my #75 back was Trent Richardson, that would pass the smell test. But if I told you my next three backs were C.J. Spiller, Darren Sproles, and Justin Forsett, and my Top 10 also included Derrick Henry, Mark Ingram, C.J. Anderson, and Reggie Bush, while Todd Gurley ranked 24th, Le'Veon Bell ranked 27th, Marshawn Lynch ranked 35th, Frank Gore was 43rd, David Johnson was 57th, Melvin Gordon ranked 59th, and Steven Jackson ranked 61st... suddenly that doesn't pass the smell test any more. Yet that's exactly what you see if you rank running backs by yards per carry. It's clearly not doing a very good job of measuring talent.
In fact, Danny Tuccito has calculated how long it has historically taken various statistics to “stabilize”— to reach a point where they are more representative of player talent than they are of noise, luck, or random chance. For instance, for Yards per Attempt (arguably the single best "simple stat" in all of football), it takes about 396 pass attempts before a player's average represents 50% skill, 50% luck. After a little bit less than a full season in an offense in we can be pretty confident which quarterbacks are pretty good and which are not based on yards per attempt alone.
For yards per carry to stabilize, a back would need about 1978 carries, (in Danny's words, "a vomit-inducing" 1978 carries). For context, that's more carries than Maurice Jones-Drew or DeAngelo Williams had in their entire career. Among active RBs, only LeSean McCoy, Adrian Peterson, and Frank Gore have reached that carry threshold, but ypc hasn't stabilized for those three backs yet because we actually need 1978 carries on the same team and in the same offense. Essentially, the practical answer to the question of when yard per carry stabilizes is “never”. A back's yard per carry is always more luck than skill.
What does it mean to say that yard per carry is always more luck than skill? Well, for one thing, the correlation between yard per carry in one year and the next is extremely low. Not only that, the correlation between yard per carry between one 8-game sample and another 8-game sample in the same season is extremely low.
If a running back averages 5.00 yard per carry in one 8-game sample, based on regression we'd expect him to average 4.37 in the other. If a running back averages 3.50 yard per carry in one 8-game sample, we'd expect him to average 3.93 in the other. Thanks to the magic of regression to the mean, a chasmic 1.5 yard per carry difference shrunk to a barely noticeable 0.44 yard per carry difference.
So like I said at the top, any discussion of regression to the mean would be remiss not to lead off with yard per carry. This is the quintessential regression stat. It doesn't really measure how good a player is, it's always more a product of luck than skill, and it fluctuates wildly and randomly between samples.
Volume, on the other hand, is incredibly sticky. Backs who get a lot of touches with a low yard per carry average are likely, going forward, to get a lot of touches with a higher yard per carry average. On the other hand, backs who get a few touches with a high average are likely, going forward, to get a few touches with a lower average.
I've made the "yards per carry will regress dramatically" prediction four times in my two years running Regression Alert, and I've tracked the results every time. So far the prediction is 4 for 4, usually paying off dramatically. (Last year, the running backs in the NFL with the lowest ypc over the first two weeks actually averaged more yards per carry over the rest of the season than the running backs with the highest ypc.)
Regression is not about being 100% right 100% of the time; if you hit on 70% of your bets in fantasy football you'll dominate your league. A day will come when I make the yard per carry prediction and it fails. But I'm happy to bet that today is not that day.
23 running backs have rushed for at least 100 yards so far this season. Let's throw out the middle three and compare the top ten to the bottom ten.
Devin Singletary, Justin Jackson, Saquon Barkley, Dalvin Cook, Carlos Hyde, Mark Ingram, Raheem Mostert, Alvin Kamara, and Todd Gurley have collectively rushed 255 times for 1645 yards in 20 total games, an average of 6.45 ypc. This is our Group A.
Derrick Henry, Christian McCaffrey, Ezekiel Elliott, Aaron Jones, Austin Ekeler, Leonard Fournette, Nick Chubb, Peyton Barber, Chris Carson, and Le'Veon Bell have collectively rushed 332 times for 1373 yards, an average of 4.14 ypc. This is our Group B.
Group B has gotten 30% more carries than Group A, but Group A is out-rushing Group B by 20% thanks to the magic of yards per carry. Since ypc is essentially just random noise, I predict that Group B's higher workload will win out and those backs will rush for more yards per game than the Group A backs over the next four weeks. As always, I'll track the progress of this prediction, so tune back in to see how well we do.
