DynastyFootball

A New Running Back Model to Rank the 2013 Draft Class

LeVeonBellMSU_thumb.jpg

Now that the NFL draft has finally come and gone it leaves us with the most valuable piece of information for projecting running back fantasy success…draft order.  To me draft order can be decomposed into two parts: Scouting consensus (how good the player is on both an absolute and relative positional level) and team need/desire.  We try to hammer home at RotoViz that the scouting model is at best inefficient, and at worst fraught with error, but let’s set that aside for a moment and acknowledge that whatever the pitfalls may be, scouting and its effect on running back draft order says a lot about how a team feels about a player and how much they’re likely to use him.

With this valuable input in hand, I set out to redo my own personal rookie RB rankings by constructing a better model.  I started off thinking more data was better (I started with 2000), but after considering how the pass-friendly shift in the league and the devaluation of running backs that has come with it has really taken place over the past 5 or 6 years I decided to narrow my sample down to RBs who were drafted from 2007 on.  I fear that a model that best fits guys who succeeded 10 years ago may not be great at projecting guys coming into the league today.

For inputs, I started out with some of the familiar stats: YPC, Speed Score, Draft Slot.  But I also wanted to see if a Dominator Rating (DR), akin to what Shawn Siegele unleashed on the world in his insightful piece here, had any relationship to success.  Although he says that “The running back position doesn’t lend itself to [DR] analysis because draftable running backs overwhelmingly dominate carries at their universities.” I still felt it was worth looking into.  While it’s true that the average DR of a college running back is about .50 (which would already be elite for a receiver), there is still a good deal of variance around that number.  I hypothesized it would still meaningful to measure how far above or below that level a tailback measures.  I’d say the argument works exactly the same as it does with receivers: if a player is talented, he should have the biggest share of the rushing yards and touchdowns on the team.  The more talented he is, the more of the share he should get as the team uses their best ground weapon to try to win games.  That’s actually the very analysis Shawn does in the piece I lifted the quote from above.

One pitfall here, of course, is the running back from a team with a mobile quarterback.  His DR will be artificially suppressed by the QBs rushing share.  And with any other season-long measure it’s subject to the effect injuries have on a player missing games.  These are all factors to keep in mind when using a model like this to do your final rankings.

Despite believing in the power of Agility Scores I stayed away from using it because I’m wary of relying too much on any measurement that isn’t always available for a prospect.  In practice, it’s probably better to move a guy with an elite Agility Score up your rankings a spot or two from where he would otherwise be.  But I have trouble moving anyone too far on a metric that only some people have.  Should CJ Spiller have ranked at the top of the 2010 running back class because of his elite agility?  We’ll never know because we don’t have his score…

Rather than trying to find a linear relationship between model scores and fantasy finish or points-per-game, I decided to take a frequency approach.  Predicting the future is very difficult, especially with something so rife with randomness as sports, so I wanted to lower the bar just a little.  Instead of trying to measure model success on predicting exactly how successful a running back will be relative to all others, I wanted to see if I could make a model that could simply weed out the guys who will hit from the guys who won’t.  Or put more simply, can my model do better than a coin flip?

I defined fantasy success by looking at the number of top 30, 20, and 10 RB finishes a player has had between 2007 and 2012.  In other words, did he produce a startable fantasy season at the flex, RB2 or RB1 level?  So, Chris Johnson (as disappointing as he has been) gets credit for 5 top 30 finishes, 5 top 20 finishes, and 2 top 10 finishes.  In order to avoid trying to fit the model to the older guys who have been successful consistently since 2007 I looked at it from both the qualitative approach:  for the top 35 players identified by the model, how many collective top 30, 20, and 10 finishes did it produce? as well as the quantitative approach (call it hit-ratio): how many of the top 35 players identified had at least one finish inside those benchmarks?  This way you can evaluate whether the model is capturing both the guys who are consistent producers, as well as the recently drafted guys who haven’t had time to string together multiple good finishes.  I used 35 because that gets you the top 5 or 6 RB prospects from every season, which is about the maximum number you can reasonably expect to develop into a fantasy relevant starters from any class.

First, let’s look at the model results for some of the individual factors:

Factor

T30 Ratio

T20 Ratio

T10 Ratio

T30 Hit

T20 Hit

T10 Hit

Avg Year

Draft Slot

1.49

1.17

0.66

60.0%

48.6%

34.3%

2009.4

YPC

1.09

0.80

0.31

45.7%

34.3%

20.0%

2009.5

Speed

1.00

0.74

0.46

40.0%

31.4%

25.7%

2009.8

DR Rush

0.94

0.71

0.37

37.1%

34.3%

20.0%

2009.5

DR Rec

0.63

0.54

0.26

31.4%

25.7%

20.0%

2009.7

As expected, Draft Slot is by far the best individual factor.  The rest are worse than a coin flip on their own.  I included the Average Year to make sure I wasn’t capturing only guys from either the beginning or the end of the time period.  2009.5 is the midpoint of the 2007 – 2012 range.

At first blush, Rushing DR on its own doesn’t even look very promising at all.  But after trying multiple combinations of several factors I stumbled upon one that works with strikingly good results.  Check this out:

Model

T30 Ratio

T20 Ratio

T10 Ratio

T30 Hit

T20 Hit

T10 Hit

Avg Year

Draft,DR Ru,Speed

1.60

1.26

0.71

62.9%

51.4%

37.1%

2009.6

It improves the results quite significantly over any of the individual factors alone.  I think in words what this model is saying is “Running backs who are successful tend to be ones who are drafted in the first three rounds.  The ones who are most successful tend to have good size/speed combinations and account for 65% or more of their teams’ rushing yards and touchdowns.”

For some anecdotal results, I love that it does better than either solely using NFL draft position or consensus rookie draft order to project the the plentiful 2008 class:

Player

NFL Draft

NFL Rank

ADP

ADP Rank

Model

T20 Finishes

Matt Forte

44

6

39

3

1

5

Chris Johnson

24

5

58

6

2

5

Ray Rice

55

7

70

8

4

4

Rashard Mendenhall

23

4

52

5

5

3

Jamaal Charles

73

9

83

9

8

3

Darren McFadden

4

1

18

1

3

1

Jonathan Stewart

13

2

35

2

6

1

Kevin Smith

64

8

39

3

7

1

Felix Jones

22

3

59

7

9

0

And now for the good stuff…the 2013 class.  The good news is there is a standout prospect who has landed in a fantastic situation.  The bad news is there is almost no one else who ranks highly enough on this model to be expected to yield much fantasy production:

Name

Team

Height (in)

Weight (lbs)

40 Yard

Speed Score

Agility Score

DR Rush

Draft Slot

Model Score

Le’Veon Bell

Steelers

73

230

4.60

102.7

10.99

0.92

48

1.43

Giovani Bernard

Bengals

68

202

4.53

95.9

11.03

0.48

37

0.69

Montee Ball

Broncos

71

214

4.66

90.8

11.28

0.57

58

0.62

Eddie Lacy

Packers

73

220

4.55

102.7

0.44

61

0.50

Johnathan Franklin

Packers

70

205

4.49

100.9

11.20

0.55

125

0.12

Christine Michael

Seahawks

70

220

4.54

103.6

10.71

0.20

62

0.07

Zach Line

Vikings

72

232

4.65

99.2

11.52

0.67

0.04

Jordan Roberts

Chiefs

70

222

4.49

109.2

11.38

0.61

0.04

Knile Davis

Chiefs

71

227

4.37

124.5

11.34

0.22

96

0.02

Stepfan Taylor

Cardinals

69

214

4.76

83.4

11.63

0.61

140

-0.07

Dennis Johnson

Texans

67

196

4.48

97.3

11.66

0.60

-0.10

Denard Robinson

Jaguars

71

199

4.34

112.2

11.31

0.39

135

-0.12

Marcus Lattimore

49ers

71

221

4.62

97.0

0.46

131

-0.12

D.J. Harper

49ers

69

211

4.52

101.1

11.42

0.56

-0.13

Miguel Maysonet

Eagles

69

209

4.54

98.4

11.64

0.56

-0.16

Robbie Rouse

Vikings

65

190

4.80

71.6

11.22

0.68

-0.21

Stefphon Jefferson

Titans

71

213

4.68

88.8

11.41

0.55

-0.26

Zac Stacy

Rams

68

216

4.55

100.8

10.87

0.46

160

-0.32

Montel Harris

Buccaneers

68

208

4.68

86.7

11.13

0.52

-0.33

Latavius Murray

Raiders

75

223

4.38

121.2

11.17

0.44

181

-0.35

Ray Graham

Texans

69

199

4.80

75.0

11.38

0.58

-0.36

George Winn

Texans

71

218

4.75

85.6

0.51

-0.36

Joseph Randle

Cowboys.

73

204

4.63

88.8

0.45

151

-0.38

Mike Gillislee

Dolphins

71

208

4.55

97.1

11.52

0.46

164

-0.39

Matthew Tucker

Eagles

73

221

4.55

103.1

0.36

-0.46

Onterio McCalebb

Bengals

70

168

4.34

94.7

0.35

-0.56

C.J. Anderson

Broncos

68

224

4.60

100.1

11.27

0.29

-0.61

Kenjon Barner

Panthers

69

196

4.52

93.9

11.07

0.43

182

-0.63

Michael Ford

Bears

70

210

4.50

102.4

11.12

0.23

-0.68

Cierre Wood

Texans

71

213

4.56

98.5

0.24

-0.71

Mike James

Buccaneers

71

223

4.53

105.9

0.34

189

-0.74

Chris Thompson

Redskins

67

192

4.42

100.6

0.18

154

-0.76

Jawan Jamison

Redskins

67

203

4.68

84.6

0.63

228

-0.77

Kerwynn Williams

Colts

68

196

4.48

97.3

11.30

0.53

230

-0.83

Andre Ellington

Cardinals

70

199

4.61

88.1

0.37

187

-0.83

Theo Riddick

Lions

70

201

4.68

83.8

0.29

199

-1.11

Rex Burkhead

Bengals

70

214

4.73

85.5

10.94

0.17

190

-1.23

Spencer Ware

Seahawks

70

228

4.62

100.1

11.34

0.10

194

-1.25

Scores over 1 signify a very high likelihood of solid fantasy production.  Scores between 0.75 and  are 0.50 are a good bet for some production, but the relationship is weaker.  Scores below .50 are lottery tickets.

A few things to note:

  • The Le’Veon Lovin’ continues.  He’s a clear-cut stud with this model.  Add in an ideal situation where he’s been drafted to compete for (or perhaps already has?) the starting job – I’m toying with putting him atop my personal rookie draft board, ahead of even Nuk.

  • Christine Michael gets whacked on his low Rushing DR partly because of Johnny Football.  If his DR were higher, based on his Speed Score and draft position, he would probably rank #2 or 3.

  • Zac Stacy was a disappointment.  I tend to buy the hype surrounding him as a good candidate to take over the StL starting job, but my model score makes me have to second guess that.  Of all the backs on their roster he does score 2nd for me, behind Pead…

  • I was curious what would happen if I plugged in Marcus Lattimore’s non-injury-marred 2010 rookie campaign stats and bumped his draft position up to late 1st/early 2nd.  He would still rank #2 behind Bell, but he ends up in elite territory with a score of 1.14 for whatever that’s worth.

  • I find it eerie how many times guys I have back to back on model scores were selected by the same team:  Lacy/Franklin, Roberts/Davis, Lattimore/Harper, Graham/Winn, Thompson/Jamison.  Probably just a coincidence, but I found it amusing.

     

Subscribe for a constant stream of league-beating articles available only with a Premium Pass.

By Ryan Rouillard | @ryanrouillard | Archive

No Comment

Leave a reply