RotoDoc holds a Ph.D. in mathematics and tackles sports data to give you a competitive edge in your fantasy games.
Last week, I created running back projections to project statistical totals for the first five weeks of the season. The goal of that exercise was to find discrepancies between my projections and current ADP in MFL10s. However, beyond simply ADP, I think there are a few other insights the model can be useful for such as identifying insights into running back by committee (RBBC) situations, PPR vs. non-PPR differences, and some opportunities for model improvement.
There are two interesting RBBC situations that I wanted to quickly touch on. First, the New York Giants (current) three headed monster. Of the three, I’m most concerned about Rashad Jennings. The projections have Jennings and Andre Williams taking carries from each other, with Jennings opportunity forecast score (OFS) at 0.35 and Williams’ OFS at 0.4, showing the model is leaning toward Williams seeing a higher rushing workload. As expected Shane Vereen plays the pass catching role, keeping him viable in PPR formats, while not a threat to vulture considerable carries from Williams or Jennings. Given that Williams is currently being drafted behind Jennings and Vereen, there’s potential value here. For more on this situation, I suggest you listen to Rich Hribar’s thoughts on Williams from last week’s RotoViz Radio podcast.
The other RBBC situation to examine is the Oakland Raiders. With the signings of Roy Helu and Trent Richardson, it appears as if Latavius Murray will see considerable competition for touches, keeping his value limited. I do believe, however, that this is a situation where the model gets it wrong in terms of the order it ranks these three. The model has Richardson projected 32nd, followed by Helu in 37th, and finally Murray in 38th. Murray should see the highest snap count, as predicted by the OFS metric, which correlates well with PPR points. When I make a few tweaks to the model down the road, I expect Murray to pop out as the highest ranked of the three. However, if it’s a choice between Murray and the guy getting drafted immediately behind him, Joique Bell, give me Bell based off expected usage alone.
PPR Formats – Hyde and Seek Someone Else
Scoring format matters, and in a PPR format like MFL10 I cannot advocate for taking Carlos Hyde at his current ADP. He is projected to have the 15th fewest receptions among the 80 running backs in the projections. When we narrow it down to guys who are expected to be their team’s lead back1, Hyde projects to have the 3rd fewest receptions of that group. Look, I know the model is ludicrously underestimating his rushing stats (the model had Hyde at 55th among running backs, which goes to show it needs some tweaks), but even if we gave him identical projected rushing stats to Lamar Miller2 who sits one spot ahead of Hyde in MFL10 ADP, that would land Hyde at 17th among RBs in my projections at 59.93 PPR points over the five games. On the flip side, Miller’s projected 72.41 PPR points shows he has about a 12.5 point advantage over Hyde just in receiving stats alone over those first five weeks. Given additional uncertainties surrounding this situation such as a new head coach3, the arrival of Reggie Bush to limit upside in the receiving game, and a tough strength of schedule4, there are just too many questions and limitations for me to recommend Hyde at an ADP of RB13 in PPR formats. That ADP seems to be closer to his ceiling than his median outcome.
Just a few quick hits on some oddities of the model outputs, with which better (and possibly more) inputs can either help correct or confirm.
- Knile Davis is projected for more rushing TDs than Jamaal Charles over the first five games…uhhhh what?
- Adrian Peterson and Danny Woodhead have been left off the projections because they did not meet the minimum carries requirement. Since I used last year’s data to project this year, I’ll have to find an alternate method to project these cases.
- Bryce Brown is projected to have very little opportunity (0.06 OFS), yet is projected to have more receptions than LeSean McCoy who has an OFS of 0.47.
- Even at an R-squared of 0.51 on predicting PPR points, you can see the spread around the best fit is quite large. For instance, if we look at the predicted PPR points value of 50, we can see the actual PPR points values range from around 18 to 80.
Some of these kinks may never be fully worked out, but I think working in better independent variables for projecting receiving stats, more accurately calculating OFS, and not having a carry cutoff should help with model accuracy. Hopefully a few tweaks will push those R-squared values for projected PPR points above the 0.6 threshold and give an even more accurate forecast for 2015.
|Rk||Player||Tm||OFS||Pred PPR Pts||Pred RuYds||Pred RuTD||Pred RecYds||Pred Rec||Pred RecTD||Pred Fmb|