By the Numbers: Forecasting RB Performance
Now that the NFL draft has come and gone, we’re left to sort out the implications for our dynasty rookie picks and redraft fantasy teams. We all know that draft position is predictive of future NFL success, but how much should we weight it versus our pre-draft evaluations? By finding the right set of relevant variables, we can develop a numbers-based formula that would have told us that Charles Sims (drafted 69th overall) was a better prospect than Bishop Sankey (54th), and Jeremy Langford (106th) looked more like past NFL successes than T.J. Yeldon (36th).
In this article our goal is to predict NFL success for recently drafted running back prospects, using collegiate production statistics, NFL combine measurables and NFL draft data. We’re going to define NFL success in a way that would likely lead to a profit on your dynasty rookie draft picks, and that’s reaching a threshold season-long finish in a player’s first three years. For running backs, we chose a top-12 finish (PPR scoring) as the most logical threshold to define NFL success.
To determine the variables that, in addition to draft position, are statistically significant for predicting early NFL success, we used a logistic regression model. The logistic regression model builds a formula for predicting a binary outcome, in this case NFL success, by weighting the relevant independent variables.
The Results
The five variables we found best for predicting early NFL running back success are (in order of statistical significance):
- Draft position (logarithmic formula to capture exponential decay of value)
- 40-yard dash time
- Receptions
- Rushing yards per game
The reception and rushing numbers are taken from each prospect’s most productive season, since some of the prospects missed significant time during their final college year.
Here are the top-15 predict scores (roughly equivalent to the likelihood of success) in the 2000-2013 data set used to train and test the model.
Top 15 Running Back Scores (2000-2013)
The top-15 backs are mostly high draft picks, reflecting the importance of draft position as a measure of talent and future opportunity. The second most important variable, forty time, boosts successes Chris Johnson and DeMarco Murray higher on the list than their draft positions alone would indicate.
Despite missing on Ronnie Brown, who had the most favorable prediction, the model correctly predicted the rest of the top-10, and even the misses in the top-15 all had productive seasons at some point in their careers.
Some might be offended that LaDainian Tomlinson and Adrian Peterson aren’t even higher on the list, considering how great they’ve been in the NFL. You have to keep in mind is that this is a predictive score, without the benefit of hindsight. Both backs were hurt by their lack of college receiving production, a negative as our top-12 threshold is measured in PPR scoring.
The 2014 and 2015 draft classes were not part of the model development due to the lack of seasoning, but it’s helpful to look at those scores anyway to see if the model has any predictive value with an out-of-sample group.
2014-15 RB Scores
* Based on low NFL Draft Scout estimate
Todd Gurley was the only running back from the last two drafts to have higher than a 50 percent chance of early NFL success, according to the model. Gurley’s predict score would have been even better if he hadn’t missed time in his final college season due to injury.
Despite being a mid-first round pick and having insanely high rushing production, the model didn’t see Melvin Gordon as a potentially elite producer. Gordon’s forty time was decent for a back of his size, but not elite by any stretch.
Two successful running backs who didn’t make the top-15 predict scores were Devonta Freeman and Jeremy Hill, both held down by poor forty times (4.58 and 4.66, respectively). I’m not sure if their exclusion was flaw in the model, or reflects flukey performances that may not be easily predicted.
The model dislikes the unathletic and relatively unproductive T.J. Yeldon, despite his early second round draft position. The Jaguars are known as a franchise that embraces analytics, but taking a running back with Yeldon’s profile that early in the draft belies their number-crunching reputation.
Now let’s move on to the 2016 draft class.
2016 Draft RB Scores
* Based on NFL Draft Scout Estimate
Dynasty ADP from RotoViz Dynasty ADP App
The hype surrounding Ezekiel Elliott may be warranted based on his studly combination of high draft position, weight-adjusted speed and strong collegiate numbers. Elliott’s 0.81 predict score is head-and-shoulders (and perhaps chest-and-torso) above the rest of the 2016 class. In fact, Elliott’s score ranks him fourth of the 400-plus running backs drafted since 2000.
Not to rain on the Zeke-love parade, but the third highest predict score – slightly better than Elliott’s – belongs to Trent Richardson. It’s important to keep in context that an 80 percent chance of success still leaves a one-in-five chance of failure. Even the greatest prospects are still that: prospects. The true test is how they perform at the next level.
Other than Elliott, there isn’t a whole lot to get excited about for the 2016 class, as most NFL teams passed multiple times on the position in the early rounds of the draft. Even the size-speed freak Derrick Henry had a fairly modest predict score, mostly reflecting his lack of receiving prowess. You should give Henry a bump in standard scoring leagues.
The pleasant surprise of the group is Tyler Ervin, whose 0.21 predict score isn’t stellar, but having a one-in-five shot at a top fantasy season is much more than you might expect considering his smaller frame, small-school pedigree and weak dynasty ADP (RB14). Ervin has the top-notch receiving ability and speed the model loves, and his rushing production is near the top of the class. Ervin’s near-term opportunity looks low behind newly signed Lamar Miller, but there should be some room for Ervin to mix in, as Miller has never functioned like a true workhorse in college or the NFL (max 227 rushing attempts).
C.J. Prosise has a lower score than Ervin, but it might be understated by the fact that he missed some action, limiting his receptions. If the converted wide receiver can adjust quickly to the NFL game, Prosise might have the opportunity to make a rookie splash.
The loser in the model is Kenyan Drake, who has a predict score of only 0.09, despite being the third running back taken this year. It’s possible that Drake’s chance of NFL success is understated in the model as he had to compete with Heisman Trophy winner Henry for touches and production. That said, it might be the prudent move to select a back like DeAndre Washington later in dynasty rookie drafts instead of Drake.
Kenneth Dixon fell further than expected in the reality draft, but the common perception of the Ravens as a prime landing spot has kept his post-NFL draft dynasty ADP in the mid first round. The destination-agnostic model thinks Dixon is severely overvalued, as backs with his combination of lower draft position and sub-par speed are rarely early NFL successes. In addition, Dixon doesn’t have as gaudy of a reception total as you’d expect based on his reputation as a top-notch passing option.
His low predict score might not seem to make Keith Marshall a target in dynasty drafts, but the model sees the speedster’s chance of success as roughly equal to players taken 100 picks earlier. Marshall, once a five-star recruit and a higher pedigreed back than fellow freshman Todd Gurley, could have had a much different career with better injury luck. The model’s modest production stats for Marshall are from when he was an 18-year-old freshman, before suffering multiple injuries that limited his playing time and allowed studs like Gurley and Nick Chubb to take the backfield reins.
