Growing up my favorite football players were the running backs. Earl Campbell, Tony Dorsett, and Walter Payton topped the list for me in the mid-1970s. And then – as a Lions fan — Billy Sims in the early 1980s and Barry Sanders in the latter part of that decade became my favorites.
We all know that Sanders was the greatest running back, both on this planet, in this universe, and in at least five parallel universes from which we have collected data.
Who, though, is the “best” running back today?
To evaluate quarterbacks we have proposed QB Score, which is calculated as follows:
QB Score = All Yards – 3*All Plays – 50*All Turnovers
Perhaps we can modify this simple formula to measure the contribution of running backs. Running backs both run with the ball and catch passes. These players also tend to fumble. Given what we know about the value of yards, plays, and turnovers, the performance of a running back can be evaluated as follows:
RB Score = All Yards – 3*All Plays – 50*Fumbles Lost
All Yards includes yards gained from rushing and catching passes. All Plays includes rushing attempts and receptions. And Fumbles Lost is how running backs lose the ball (except for the occasional errant half-back pass).
Yahoo.com reports data on all these statistics back to 1994 (I can’t find fumble lost before 1994). With RB Score in hand, I evaluated all running backs from 1994 to 2006. The top 40 in RB Score are listed HERE.
When we think of top running backs we typically focus on yards rushing. Backs, though, gain yards via the passing game as well. Plus, it’s important to note that fumbles lost also impact outcomes. When we consider more of what a running back does on the field (we don’t have everything since we are not considering such things as blocking or dropped passes), our list of top running backs differs from what we see when we just look at rushing yards.
Since 1994 the best rushing performances was posted by Jamal Lewis, who gained 2,066 yards on the ground in 2003. He also lost six fumbles that season – a mark that tops all of the Top 40 running backs listed – and only gained 205 yards receiving. So his overall productivity – measured via RB Score – only ranks 32nd.
Lewis posted a RB Score of 732 in 2003. This is good, but a bit short of the 1,ooo mark that appears to separate the very good from the truly outstanding. Since 1994 only six running backs have posted an RB Score above 1,000. Tiki Barber (in 2005), Priest Holmes (in 2002), and Charlie Garner (in 2002) each did this once. Marshall Faulk turned the trick twice (in 1999 and 2000). Barry Sanders also did this twice, in 1994 and 1997. Finally, LaDainian Tomlinson has done this twice. The first time was in 2003 and the second time was this past season. So according to RB Score, Tomlinson was the best back in 2006 (which is not a surprise).
A mark of 1,000 is relatively rare. As one can see in the top 40 list, a mark close to 700 is still quite good. In 2006, six backs not named LaDainian came close to 700 or moved well past this benchmark. These six backs include Steven Jackson, Tiki Barber, Brian Westbrook, Frank Gore, Larry Johnson, and Maurice Jones-Drew.
The overall RB Score is interesting, but one might also wonder who gives you the most per times touching the ball. If we look at RB Score per Play, which one can see HERE, the top 40 looks a bit different. Since 1994 only four running backs have had a per play mark above three. These include Faulk – who did this twice – as well as Sanders, Garner, and from 2006, Jones-Drew.
Does this mean that Jones-Drew was the best back in 2006? For running backs there is value in durability as well as efficiency. Jones-Drew has yet to show that he can be productive getting 300+ carries in a season. Hence overall RB Score might be more indicative of value.
The Top 40 lists tell us about the best of the best in the last 13 seasons. What about the 2006 collection of running backs? If one looks HERE, one can see the 52 running backs who participated in at least 100 plays this past season. The top seven backs have already been noted. Just off the very top we see two backs from New Orleans. Reggie Bush was ranked 41st in rushing yards this season, but 11th in RB Score. Deuce McAllister was 20th in rushing yards, but 15th in RB Score. Coupled with the performance of quarterback Drew Brees, we can see how the Saints improved so much in 2006.
The Saints backs were among the best. What do we see when we look at the worst? If we look at the end of the list we see a surprising name, Shaun Alexander. Alexander’s 2005 performance appear on the top 40 list of RB Score. In 2006, though, his final RB Score was only 2. That’s not his ranking, that’s his total RB Score. So that’s quite a drop off.
Inconsistency is not unique to Alexander. As has been noted in this forum, quarterbacks are inconsistent. Less than 20% of a quarterback’s QB Score per play can be explained by what the quarterback did last year. For running backs it’s the same story. When we look at running backs we again see that less than 20% of current performance – measured with RB Score per play – can be explained by what the running back did last year. And this is the same story we see when we turn to the metrics for running backs developed by Football Outsiders.
For Alexander, injury seems to have impacted his performance. But we can imagine that injuries are not the story for everyone. Edgerrin James offered an RB Score per play of 1.44 with the Indianapolis Colts (average is 1.1). With the Arizona Cardinals his per play performance declined to 0.27. Certainly one suspects that it was the talent difference between the Colts and Cardinals that contributed to the drop-off in the productivity of James.
The list of best and worst is fun, but the consistency story is the real tale to be told. Like quarterbacks, a running back depends on the players around him to perform. So when we look at each back’s stats we are not seeing a measure of that player’s talents, but also the talents of his offensive line (and quarterback, receivers, and coaching staff). In sum, stats in football – unlike baseball or basketball – do not measure individual performance very well. Of course, that story should not detract from the fun we have looking at these numbers.
– DJ
Wages of Wins Journal 
Daniel
January 15, 2007
An interesting note from the Eagles – Sanits game … Total RB score per play between Saints runners (Bush & McAllister) was 3.37 … Total RB score per play between Eagles runners (Westbrook & Buckhalter) was 4.88 … Perhaps the Eagles should have run Buckhalter more?
RSaunders
January 15, 2007
It’s interesting that Emmit Smith doesn’t have even one season ranked in the Top 40. I suppose that makes sense since he seldom racked up huge yards or caught many passes. But, neither did he fumble very much. I would’ve expected more Ricky Watters, too. Is it essentially the fact that it took them more plays to produce their count statistics (rushing yards, receiving yards)?
vijay
January 15, 2007
Is there any way to control for variable offensive line talent?
Jason
January 15, 2007
One problem I’ve had with football stats is that football, far more than the other major sports, seems to have more situational dependencies in addition to being far more dependent on the performance of teammates. A running back sent out in deep patterns will have a higher yards per catch average than one sent on shallow routes. There is no penalty built in for the dropped pass or the blown route. The back who catches the outlet under for 3 yards, perhaps preventing a loss when his QB would otherwise be sacked, dilutes his score, though he may well be more helpful than the one who dropped the same pass (or wasn’t open enough to justify even getting the ball in his direction).
Clearly, similar arguments are made in basketball and football. I’ve heard what seem like never-ending diatribes about ‘garbage time’ points and the batter who seemingly only hits when his team is up or down big. I strongly suspect that these are rarely that important in the grand scheme of things, but I would be far less surprised if situational variables, including the talent of teammates and the schemes instituted by coaches, wasn’t exceptionally important in football.
dberri
January 15, 2007
RSaunder,
My study only goes back to 1994. That year Emmitt Smith’s performance ranks 67th. His performance the next year ranked 81st. After that his highest mark is 176th. I looked at all running backs with at least 100 attempts from 1994 to 2006. There were 553 backs in the sample, so 176 is still above average.
vijay,
I do not have data on the quality of offensive lines. Football Outsiders had done some work in this area. Not sure how to integrate what they have done with what I have done (haven’t thought about yet, either).
Jason,
I think data on football should be summarized “these numbers are for fun, not for analysis.” The interaction between teammates in football is, I think, a huge issue. I would argue that in basketball this is a much smaller issue and in baseball I don’t think teammates matter at all (except for pitchers).
John DePalma
January 15, 2007
I think this measure is overvaluing running backs with a lot of receptions.
The part of “all yards” that reflects rushing seems fine. I like how it benchmarks rushing yardage against attempts and a hurdle for how much each attempt should generate. Similar to how the NBA measure reveals the tendency of traditional statistical metrics to overvalue scoring, this method is showing how total rushing yardage is overvalued. The subtraction for lost fumbles seems mostly apt also, but I have a cosmetic objection. A bunch of statistical work reveals how fumble recoveries are mostly random so perhaps the better measure would be total fumbles. Then multiply total fumbles by some number consistent with the average % of fumbles lost.
But I think the treatment of receptions could be very problematic. In 1999 Marshall Falk, as part of the supposedly best running back season of the past decade, had 87 catches for 1,048 yards, or 12.0 yards/catch. He also had 253 rushes for 1,381 yards, or 5.5 yards/rush. And he fumbled twice, losing both fumbles.
His RB score per play is 3.85. Ignoring interceptions the RB score per play is 9.0 on pass plays and 2.5 on rushing plays. I would argue that receptions represents a selection bias. If the ball was thrown to the running back 5 times for each 3 receptions (i.e. 60% completion %), then perhaps plays in the RB score should be scaled upwards by about 70% to correct for this bias.
dberri
January 15, 2007
John,
I don’t disagree. It might be the case that Jamal Lewis simply suffered from bad luck when he lost that many fumbles. And it would be nice to know how many times running backs were thrown to.
At the end of the day, though, (as I noted to Jason) these measures are mostly for fun. I just don’t see how to untangle what a back does from his offensive line and other offensive talent. So no matter how you do RB Score (and what you suggest seems reasonable), this is not going to be a perfect reflection of how “good” a running back is. In sum, despite the numbers, I still think Barry Sanders at his best was better than Marshall Faulk at his best.
dylan
January 16, 2007
I agree that something seems to be weird in that receiving running backs are getting placed to high, and am curious what the list looks like if you include intended passes that were incomplete. I think this data is pretty easily available, and would be a simple and easy improvement on the stat as the best running backs are going to be the ones that not only catch a lot of passes for a lot of yards, but also a high percentage of passes to keep the offense out of bad situations.
dberri
January 16, 2007
Dylan,
If you have data on dropped passes by running backs, please send it along. There is a problem with RB Score because you only get credit for completed passes, and not charged for dropped passes.