In Part I, I looked at the overall relationship between workload and age. In that post, I focused on only the top backs of the last 25 years; today I want to expand the sample and look at production on a year-to-year basis.
Last year, Joique Bell finished as the 15th best running back in fantasy football. Prior to 2013, Bell had just 82 career carries, all of which came in 2012. Meanwhile, Marshawn Lynch finished as RB5, but he had 1,452 carries prior to the 2013 season. Both players were 27 years old last year, but they had drastically different career workloads.
One obvious issue that comes up when comparing high-workload to low-workload players is that there is often a large talent gap, and Bell and Lynch present that quite clearly. Bell was an undrafted free agent out of Division II Wayne State, while Lynch was a first round pick who played in the Pac-10. What I'll try to do today is control for "player ability" by looking at the player's VBD in the prior season. For example, Lynch had 125 points of VBD in 2012, while Bell had 0.
From 1988 to 2013, there were 77 running backs who had a top-24 finish during their age 27 season. One thing we can look to see is whether these players "benefited" from having low mileage up to that point in their careers. I performed a regression analysis using three inputs -- Carries in the player's age 26 year (for example, 315 for Lynch), his career carries as of the end of his age 26 season (1,452 for Lynch), and his VBD in his age 26 season (125). My output was VBD in the player's age 27 year. Here was the best-fit formula:
Projected VBD = 43.2 + 0.023 * Career Carries - 0.091 * Age 26 Carries + 0.52 * Age 26 VBD
The key numbers here are the coefficients on the two carries variables. The evidence here suggests that having more career carries through age 26 leads to more fantasy points in age 27 (and we've tried to control for quality by including Age 26 VBD here), but that having more carries at age 26 leads to fewer fantasy points at age 27.
By far the biggest predictor in age 27 VBD is age 26 VBD, but we can look at the individual results to see what's driving the regression. Two of the biggest overachievers at age 27 were LaDainian Tomlinson and Adrian Peterson, and both had huge career workloads through age 26. Ahman Green, Robert Smith, Edgar Bennett, and Willie Parker were the four biggest underachievers -- those backs generally had big workloads at age 26, but not necessarily a lot of work prior to then. As a result, you can see why the regression would be pulled towards valuing players with large career workloads but against players with high workloads at age 26.
Of course, the sample size is not very large, and this represents just one age. We can do this for age 28, too. Here, we have 66 running backs who had a top-24 finish at age 28. What does the regression tell us?
Projected VBD = 50.3 - 0.014 * Career Carries - 0.035 * Age 27 Carries + 0.60 * Age 27 VBD
Here, we see the coefficient on both carries variable is negative, which would support the idea that a heavy workload is bad for a running back. So what's driving these results? Priest Holmes came out of nowhere to have a monster season at age 28, so that's a point in favor of low mileage. Eddie George and Edgerrin James had monster workloads throughout their careers, and both saw their VBD values dropoff sharply during their age 28 seasons. Emmitt Smith, Jerome Bettis, and Adrian Peterson also saw dropoffs at age 28; as a result, we see that the regression is a bit biased in favor of lower-workload backs.
Here's the regression for age 29:
Projected VBD = 23.4 - 0.005 * Career Carries + 0.105 * Age 28 Carries + 0.30 * Age 28 VBD
Here, age 28 carries are a good thing. Priest Holmes and Barry Sanders had monster fantasy seasons at age 29, and both had over 300 carries at age 28. Finally, here's the age 30 regression:
Projected VBD = 62.2 - 0.017 * Career Carries - .077 * Age 29 Carries + 0.776 * Age 29 VBD
Let's review what the coefficients were for these three variables for the regression at each age. In addition, I've provided simple and weighted averages of the coefficients, with the weighted averages based on the number of players in the sample.
| Variable | 27 | 28 | 29 | 30 | Avg | Wt Avg |
|---|---|---|---|---|---|---|
| Career Carries | 0.023 | -0.014 | -0.005 | -0.017 | -0.003 | 0.000 |
| N-1 Carries | -0.091 | -0.035 | 0.105 | -0.077 | -0.024 | -0.030 |
| N-1 VBD | 0.519 | 0.604 | 0.301 | 0.776 | 0.550 | 0.534 |
| # RBs | 77 | 66 | 49 | 33 | 0 | 0 |
The career carries variable seems to hold little predictive power. In fact, the analysis here suggests that cumulative workload isn't a thing. Or, perhaps more accurately, it suggests that to the extent cumulative workload is a thing, it's offset by the fact that a large number of career carries is so strongly correlated with quality, even if the player is coming off a down year.
The carries variable in the prior year is associated with fewer points of VBD in the current year, but the effect is pretty miniscule. Giving a player an extra 150 carries in the prior year only reduces their expected VBD in the current year by 4.5 points. That's less than one touchdown, so it's a very small effect to the extent it even exists. It's certainly possible that the results could have looked this way by chance. Again, untangling workload history from quality is close to impossible, so these results could still be biased due to the issue. But this is another study that doesn't seem to support the idea that a high workload in the past is a significant issue for fantasy players to worry about. That's good news for Marshawn Lynch fantasy owners.
