The Workhorse Metric: Your Tool For Scouting Fringe RBs

Last summer, I introduced the non-Quarterback Dominator Rating (nQBDR), which was an attempt to quantify (using market share) the extent to which any given runner dominated the non-QB rushing production on his team.

While I’m also a fan of raw production and don’t think that nQBDR is the only way rushing production should be measured, I do believe that nQBDR is extremely useful. This metric pointed to the potential of such late-round and undrafted prospects as Alfred Morris, Ahmad Bradshaw, Darren Sproles, Joique Bell, LeGarrette Blount, Danny Woodhead, and BenJarvus Green-Ellis, all of whom had at least one collegiate season with a nQBDR close to or higher than 90 percent. And last year, I used nQBDR to highlight Bobby Rainey, Lance Dunbar, and Chris Polk as undervalued second-year players and Bilal Powell, Jacquizz Rodgers, and Jordan Todman as undervalued third-year runners. Clearly, not all of these guys had fantastic seasons—but all of them did have moments of utility last season, which is more than almost anyone predicted for them. In general, nQBDR has been a useful tool in helping me find ignored running backs who have the potential to become NFL contributors if given the opportunity to play.

But nQBDR is not without its problems. For instance, even though nQBDR factors out games against non-FBS opponents and games in which a runner is injured, I still believe that too many “unrepresentative” contests make their way into the metric. What about all of the blowout victories against FBS opponents in which third- and fourth-string RBs see significant game action? Are those games representative of a workhorse runner’s true place in his offense? Or what about blowout losses, maybe games in which a third-string RB scores a last-minute TD only because the starter was pulled early in the fourth quarter with his team down by 35 points? Are those games representative?

After combing through data, I’ve formed the theory that whatever rushing production occurs in a blowout significantly risks being unrepresentative in its market share distribution—for both teams—whether that production occurs at the beginning of the game when the score is 0-0 or at the end of the game when the score is 72-17. Intuitively, this makes sense. A coaching staff will not use its primary RB in a 45-10 contest the same as it would in a 13-10 game, and if a team ends up winning or losing by 35 points then the manner in which it used its RBs on the way to that outcome is not likely to have been representative, because “valuable” players would have been removed from the game to avoid injury and their usage before the removal likely would not have been in accordance with the original game plan. It’s not that blowouts always render unrepresentative distributions of rushing production, but they often do—and what I really want to know is the kind of production various runners accumulate when they’re used the way that their coaches want to use them in normal, at least pseudo-competitive contests.

As a result, I’ve revamped the nQBDR metric—and I’ve given it a new, less amorphous name while I’m at it.

Ladies and gentlemen, I give you the Workhorse Metric, which is basically nQBDR with more precision.

Like nQBDR, the Workhorse Metric eliminates any rushing production that comes from a quarterback (or punter or kicker), and it also eliminates any game in which a player does not actually play. Also like nQBDR, it eliminates any game in which a player is injured and forced to quit playing if the inclusion of that contest negatively impacts the runner’s Workhorse Score (WS). I think it’s pointless to count a partial game against a RB and equally pointless to remove a game from a RB’s sample if it isn’t truly unrepresentative. Thus, some flexibility is built into the metric when it comes to games in which a runner suffers an injury.

Additionally—and this is where the Workhorse Metric differs substantially from nQBDR—any game in which a player’s team wins or loses by at least 28 points is removed from data set on which the calculation is based. Why 28 points? I had to draw the line somewhere, and that number, when I was looking at market share distributions of rushing production, was a place where the trend of un-representativeness was clear.

And one quick note about sample size. I was kind of surprised to find that, in this metric, a small sample size isn’t always the problem it might seem to be. Let me explain. If a RB plays in only six games and two of those games are factored out, then that remaining sample of four games is somewhat problematic, because it is pulled only from a half season’s worth of games at the outset. But if a RB plays in 13 games and nine games are factored out—because his team blows out almost every team they play—then his remaining sample of four games is strong, despite its small size, because that sample is still pulled from a full season’s worth of games. That sample is still representative of how he is used in competitive games over the course of the entire season, whereas the first sample risks being unrepresentative, as we don’t know how that runner would’ve been deployed in the games he has missed. So, you know, basically the difference between Cedric Peerman in 2007 and Montee Ball in 2011.

Finally, not all similar Workhorse Scores are the same. For instance, a 90 percent WS based on a final sample of 10 games is more significant than a 90 percent WS with a four-game sample. Additionally, a 90 percent WS that accompanies a season of 1,800 yards rushing is better than a 90 percent WS in a 1,000-yard rushing season. And a 90 percent WS by a RB who also catches 50 passes is better than a 90 percent WS by someone with only five catches. These are simple mental leaps.

But what about an uneven comparison? What about something like this?








MS Att



Tot WS

WS Gms

Player A












Player B












Both of these guys are 2014 rookie RBs, and these are the last years in which they played essentially a full season. (Player A missed only a couple of games in 2013; player B missed about half of the 2013 season.) As you can see, Player A—on fewer carries in fewer games—has significantly more yards rushing and TDs. By raw production, Player A seems clearly superior.

And yet when you factor out all of the non-competitive and/or non-representative games, you see that—on a similar percentage of his team’s carries—Player B dwarfs Player A with a much higher Workhorse Score based on many more games. In short, Player A has the profile of a statistical bully, scoring all of his TDs in non-competitive victories, whereas Player B served as his team’s regular TD-maker in contested games, when TDs actually mattered.

So which is more significant? The raw production or the Workhorse Score? Let me put it this way: The drafted and explosive Lache Seastrunk (Player A) got beat out for a roster spot just a few days ago by the undrafted and unathletic Silas Redd (Player B). For one season at Penn State and one season at the University of Southern California, Redd was his teams’ workhorse. For his two seasons at Baylor, Seastrunk wasn’t. Lots of factors went into Redd getting a roster spot instead of Seastrunk—but it’s exactly what the Workhorse Metric would’ve predicted.

Would you have predicted that?

OK, perhaps I should pump the brakes and offer a few words of caution: I’m still in the process of converting nQBDR to WS for a lot of players, particularly the guys who played years ago, and so I can’t say with definitiveness just how predictive the Workhorse Metric actually is over a large data set, but I can say that, with the data set I have currently, the trend of high-WS runners doing well and low-WS runner not is . . .  “intriguing,” shall we say? With the season upon us, I was faced with the decision of either 1) posting information based on work that is in progress for the sake of putting an actionable idea out there and drawing your attention to some potential stars or 2) posting nothing, continuing to build the database, and then finally posting something months from now when everything is finalized. I opted for choice #1.  It at least has the benefit of putting information in your hands, which is really what we’re all about. So, again, caveat: You’re dealing with data that is still being created and manipulated. Buyer beware.

Having said that, I also want to say that the Workhorse Metric could become an important production metric for the RB position, especially for the analysis of rookie runners. And as was the case with nQBDR, it’s helped me uncover ignored RBs who have the potential to become NFL contributors if given the opportunity to play. In other words, it helped me make the playoffs last year in all of my leagues and win a couple of championships.

If you want to see what the Workhorse Metric has to say about some 2014 rookie RBs, check out the first companion piece to this post.


Matthew Freedman is a regular contributor to RotoViz and is (not) the inspiration for the character in The League who shares his name. He serves as RotoViz’s (un)official ombudsman in the series The Dissenting Costanzan, and he also co-hosts the RotoViz Radio Football Podcast and writes The Backfield Report and The Wideout Report. He is the creator of the non-Quarterback Dominator Rating and now the Workhorse Metric and is the chief proponent of the RBx6 draft strategy and the No. 1 fan of John Brown, the Desert Lilliputian.