By Brian Burke
A couple months ago, I took a look at the safety position and how free agents are paid. And more recently I took a look at how QBs are paid with the intent to eventually establish a rational framework for how they should be paid according to their actual production in terms of wins and points. Like the prior analyses, my primary goal for now is to get a feel for the market.
As in all individual player analysis, it's worth stating up front that football is always a team sport. Like all other positions, a RB's stats are not his alone, as they depend heavily on offensive lines, opponents, scheme, situational variables, passing game strength, and other considerations. We can account for the situational variables with the EPA and WPA models, but we must rely on the tendency for factors beyond the control of the individual player to even out in the long run. They may not even out for any individual RB, but in aggregate, the idea is that some RBs will have above average lines, some will have poor lines, and collectively we can roughly estimate the financial value that a RB brings his team based on production.
First, let's take a look at how RB salaries are actually determined. There are two graphs below. The first plots career total rushing yards vs. career total cap hit. The second plots career Yards Per Carry (YPC) vs. career cap hit. The data include years 2000-2009 and is limited to RB seasons with greater than 6 game appearances and greater than 50 carries. Cap hit is adjusted for NFL inflation. Only players who had a cap hit greater than $1M in year 2000 dollars were considered. This limitation is so that we sample only high-priced RBs, and is necessary to see any relationship at all between performance and pay.
It appears RBs are paid more for total rushing yardage than per play yardage. They're paid about $2,887 per rushing yard, or if we want to throw receiving yards into the mix (which only increases the correlation from 0.89 to 0.91), we get $2,289 per total yard. This would mean a player that could reliably be expected to produce 2,000 total yards would be worth about $4.5M per year in 2000 cap dollars, which is equivalent to $9M in 2011 dollars. That's about the top of the current market for RBs.
But the method I used above aggregates career total salary and yardage, which would tend to reward past performance with future pay and obscure what's truly important--which is current performance based on current pay. So here are the same two graphs, but this time each player's season is broken out as an individual data point.
The relationship isn't nearly as strong, as the correlation drops from 0.9 to 0.3. This is what we'd expect with smaller sub-samples of attempts (divied by year), and the overall slope of the relationship drops as well. Instead of $2,887 per career rushing yard, it becomes $945 per yard. But now there is an intercept to consider. Among high-priced RBs, the relationship is $1.7M + $945*rush yd. This would suggest a RB who could be expected to produce 1,500 rushing yds would be worth only $3M compared to $4.3M in 2000 cap dollars using the aggregate model above. If we want to look at our 2,000 total-yard RB from above, the relationship would be $1.8M + $696*tot yds, for a total of $3.1M instead of $4.5M.
But as I've harped on previously, total yardage is a not a good way to measure RB effectiveness, even in the aggregate. Total yardage is primarily due to attempts, which is beyond the control of the player and usually a function of game situation. So next let's look at how advanced stats relate to salary.
The first graph plots Expected Points Added (EPA) per Play by adjusted cap hit. The second graph plots Win Probability Added (WPA) per Game by adjusted cap hit. And the third graph plots Success Rate (SR) by adjusted cap hit.
You can immediately sense that the relationship between current production, either in terms of points or in winning, is not strongly related to current salary, at least in terms of current year cap hit. (In fact, when you aggregate career stats and pay, as I did in the QB analysis and as I did above for yardage production, the relationship makes even less sense. The more a player is paid over his career, the worse his performance.)
I thought that perhaps the randomness of lost fumbles was contributing to the apparent disconnect between salary and performance as measured by EPA and WPA. If coaches and GMs understood how random fumbles are, perhaps they are properly discounting that when paying RBs. But after removing all fumble plays from the data, the relationship between pay and performance barely budged.
If I were reading this analysis, the first thing I would think is that these "EPA" and "WPA" stats must be bunk. GMs know what they're doing and they pay for total yardage, not some kooky made-up stat like EPA. The second thing I would think is that RB performance is so dependent on teammate performance, namely that of the offensive line, and on other situational factors, that it's not fair to even have "individual" stats for RBs.
I'd counter that the EPA and WPA stats fit just fine with other positions, including QBs and safeties. Further, although I agree RB performance is highly dependent on things like line play, so is passing performance. In fact, the case can be made that QB stats would be more dependent on teammates than RB stats. QBs rely on both their line and their receivers, while RBs depend (almost) exclusively on their line. The criticism that stats don't capture a RB's pass-blocking ability would be valid, but I don't think it explains the disconnect. (Does anyone believe Cleveland gave a 29-year old Jamal Lewis $10 million guaranteed over three years in 2008 because he was such a great pass blocker? Or was it because he had gobs of career rushing yards?)
I'd also say that if running outcomes depend so much more on other factors beside RB ability, then why pay high salaries to any RB at all, regardless of how good he is thought to be? Wouldn't a GM prefer to figure out what those factors are and put his limited resources there?
Either the running game just isn't as important as coaches believe or its outcomes are much more random than the present salary scale suggests. Among "qualified" players (the ones that have enough carries that their stats aren't excessively noisy), the difference between the 25th and 75th RBs in career EPA/G is 1.3 points per game. But among QBs, the difference between the 25th and 75th percentile players is 4.7 points per game.
So what kind of inferences can we make? Keep in mind nothing here is solid; it's all first cut stuff. But I think these numbers support several ideas.
First, GMs pay RBs based on rushing and total yardage, which is driven predominantly by the number of carries or touches. These are beyond the control of the RB. It could be that coaches and GMs evaluate RBs poorly, or at least not as well as they do QBs. Despite the claims that insiders don't put much weight in stats, they apparently do. They just use the wrong ones, like total yards. I hear a lot of nonsense about how so-and-so has good "vision," "instincts," and "ball skills," but ultimately pay comes down to past yardage totals.
Second, the very top RBs are generally overpaid. Whether you believe RBs have much control over their individual statistics or not, the present relationship between pay and performance is either barely existent (in the former case) or unwarranted (in the latter case). The point here is that GMs and coaches how important it is to have have a top RB on the roster. RB outcomes tend to be dominated by situational and team factors. Compared to QBs, top RBs are relatively inconsistent from year-to-year in terms of true production (as opposed to total yardage), and the importance of the running game in general is less than generally believed.
As it stands today, the league's very top RBs consume nearly 10% of their team's cap space. I believe this is too high, perhaps by a factor of 50%. More work would need to be done for a solid estimate, and it ultimately may be impossible because we'd have to untangle the relationship between the RB and his line. But based on the relative importance of the running and passing games in terms of WP, and based on the relative inconsistency of year-to-year RB performance, 50% intuitively seems like a good first cut. Instead of commanding say $9M per year, they should more rationally get about $6M per year.