Print

RB Collapses - Workload

You are viewing free content provided by FantasyGuru.com. Why not consider subscribing today?

by Mike Horn, Staff Writer

Published, 5/21/14

 

This is just a short addendum to the previous RB collapse article. You really need to read that before this one. I’m not going to repeat the study framework and definitions from that piece.

 

I know some of you were wondering if I should have focused more on the previous workload of the running backs I looked at. I’ve done some workload studies before, and believe that it’s relevant but not as important as age. But I ran a few numbers to see if previous workload helped predict RB collapses.

 

I defined “previous workload” as cumulative career touches (carries plus receptions). That’s probably not perfect as targets and blocking take a toll too, but it’s what most of us think of as “workload,” and it’s easier to gather.

 

I grouped RBs by 500-touch increments based on their workload through the end of the previous season, i.e. when they finished in the Top 10 and when I measured their age. Then I looked at whether or not they collapsed in their future seasons.

 

Did the RB Collapse?

Previous workload (touches)

Yes

No

Pct Yes

0-500

12

34

26%

501-1000

21

43

33%

1001-1500

9

28

24%

1501-2000

6

16

27%

2001-2500

3

15

17%

2501-3000

3

7

30%

3001-3500

4

2

67%

3501-4000

1

1

50%

Overall

59

146

29%

 

For example, RBs with 0-500 touches subsequently collapsed at a rate of 26%. That’s pretty close to the overall average. Most of the groups of RBs collapsed at roughly the same rate as the overall population.

 

If there is a magic number, it’s at the 3000-touch mark, but frankly those guys are pretty old too – between 27 and 31, with an average age of 29.2 years. Crossing the 2000-touch threshold actually is a positive sign, with only 17% of RBs in the 2501-2500 group collapsing. I think this is because an RB has to be both good and durable to get and sustain that kind of workload in the NFL. Of course, these observations on the 2000- and 3000-touch thresholds are based on small samples. If I rolled up the under-2000 vs. over-2000 numbers, I’d see a 28% collapse rate for the lower touch group and 31% for the higher workload RBs. Not a huge difference.

 

My main conclusion: don’t worry too much about workload as a predictor of a coming collapse. Remember that this study only looked at Top 10 backs, and that collapse is more important to dynasty owners than in re-draft formats. Finally, there is a price point (auction cost, draft value, or trade) that justifies additional risk in acquiring older RBs.

 

The next part of this is geekier and more math-heavy. You were warned.

 

Out of curiosity, I ran a regression with age and workload as the inputs, and value remaining as the output. Remember, this is for RBs who ranked in the Top 10 in a season since 1988 and who ended their careers before the 2013 season. The result was this formula:

 

Value remaining = 3260 – 116 * age + 0.12 * workload

 

The R-squared = 0.34 for this regression, roughly meaning that about 34% of the value remaining can be explained by this formula, that is, by age and workload alone. That is not a huge number, but of course there are many variables like offensive system, quality of QB and offensive line, injuries, etc. that are left out of this formula. This is actually a pretty high R-squared for a football regression based on limited inputs. That doesn’t necessarily make it good, but it has some value.

 

It also ignores a key factor that we know has a high correlation with the next season’s performance: last year’s fantasy points. We know last year isn’t everything, and it’s important to understand what can cause that number to change; that’s the point in trying to predict collapses.

 

What the regression equation also says is that every year of aging reduces an RB’s future value by 116, or roughly an RB10 finish. Meanwhile, every 1000 touches adds about 120 to future value. The first part makes sense; the future value of an RB should go down as he ages. But why would future value increase as workload also increases? Shouldn’t future go down as wear-and-tear goes up?

 

I think the answer is that workload is a proxy for both the quality and durability of a running back as well as for his wear-and-tear. In other words, a back who gets a lot of touches is generally a good back. An RB who continues to be able to play after taking a pounding is generally a durable back. A fluke injury can take out any player’s knee or end a career, but some athletes just take wear-and-tear better. Overall, we would expect a better (and more durable) RB to have more future value.

 

Age is a proxy for the same things as workload – you don’t get to be an old RB by being bad or fragile and you don’t get to be old without some wear. The fact that future value and age are negatively correlated (value goes down when age goes up) while future value and workload are positively correlated (move in the same direction) tells me this: age is a better proxy for wear-and-tear and workload is better for measuring value. Of course there could be better direct measures or proxies for any of those factors; that could be part of why R-squared is not higher.

 

Digging into the formula, an RB in the abstract starts out with a future value of 3260. Of course, the RBs in this study are at least 21 at the end of their rookie year. So they immediately drop to an average future value 824 (3260 – 116 * 21). At age 28, they only have 12 points of value left. But of course workload has to be added into that. For example, when Marshall Faulk was 28 and the #1 fantasy back, he had 2703 career touches. On average then, that would add another 324 points of value, or a total future value of 336 as predicted by the regression equation. In fact, he only produced 114 points of value and met the definition of any RB who collapsed. Fellow 28-year-olds in the same workload category (2501-3000) were Curtis Martin (who was RB5 that year, predicted value 332 vs. actual 322 – pretty good call); Barry Sanders (RB5, also predicted at 332 but his actual was 411); and LaDainian Tomlinson (RB11, 350 vs. 252). Note these three did not collapse at this point.

 

Tomlinson never did, fading more like Emmitt Smith did. Martin collapsed after his age 31 season; he was RB4 and then only produced 14 more points of value in his career. Sanders was so great he met the definition of collapse twice. In 1997 he was RB1 (age 29, value = 300). He then only produced a future value of 111, a difference of over 100, i.e. a collapse. But he generated all that value as RB10 in 1998 and then nothing more, so that counted as a collapse again. Of course, formulas and regression and collapse rates can’t predict behavior like Sanders walking away from the game. But just as Sanders sensed he wasn’t the player he had been and was unwilling to pay the price to continue as a lesser player, these numbers help to sense when an RB’s career is close to the end. The numbers will miss on some individuals and hit on others, but in the long run they will help you make better judgments.

 

You're going to score 4,610 points this week!

Back to the top