Aug 12

Running back draft strategy: controlling for outliers when assessing value

Doug Martin's historic output against the Raiders was the impetus for an exercise that considers RB fantasy value when the big games are removed (Photo: Marcio Jose Sanchez/AP).

Doug Martin’s historic output against the Raiders was the impetus for an exercise that considers RB fantasy value when the big games are removed (Photo: Marcio Jose Sanchez/AP).

Malcolm Gladwell made the topic trendy but outliers have been the bane of a statistical analyst’s existence all along. An outlier is essentially a number so far from the average (mean) that it seems out of place. In fantasy football we’re talking about statistical outputs that stand out from a player’s typical production.

For example, Doug Martin‘s 272 total yard, four touchdown performance in week nine against the Raiders; or for that matter his 33.4 fantasy point output the week prior. Anyone that was watching the game was blown away by Martin’s speed, vision, and Oakland’s tackling inefficiency. However, when you look at the numbers things just don’t add up. These outliers are performances that are hard to predict, and while Martin’s 50+ points in week nine was likely enough to win your weekly matchup, no one was predicting it and an argument can be made that it shouldn’t be considered when looking at his fantasy value for 2013. As such, it’s worth removing outliers when considering running back draft strategy overall.

Martin finished just ahead of Arian Foster in overall fantasy scoring last season as the #2 Running Back – but without his two standout efforts, he wouldn’t have been close. Foster was more or less steady – he had no monster games and very few underwhelming efforts either (except his 0.9 point effort in your fantasy championship week). Arguably, and without considering any other contextual factors, that makes Foster a better play. The big weeks are nice, but steady week to week output is really what you’re looking for from your RB1.

Weekly point outputs of 3, 5, 50, 6, 6 likely gets you a win in one week but puts you behind your opponents RB1 in four other contests. The average here is 14 points per contest. On the flip side, a back who scored 14 points week in and week out is dependable and is a weekly must start.

With that in mind, I took our top 12 consensus ranked running backs and removed any output that was more than ten points greater than their season average (replacing that number in the adjusted formula with their per game average number). I also tallied their ‘duds’ – games in which they scored fewer than six fantasy points.

The full data is here (Excel), do with it what you will (and, please feel free to share your comments below). My analysis follows below. It’s worth noting that unless a player was explicitly underused in week 17 those numbers were factored into the calculations.

[This article will be featured in the next update to our Comprehensive Draft Guide]

  • If you watched Adrian Peterson last season you’d probably come away thinking that anything is possible and that even his outliers could be considered repeatable… still even when controlling for two 31+ point outputs Peterson’s per game average is still the best in the league.
  • As mentioned above, Foster takes the edge over Martin when you control for Arian’s 0 outlier performances and Martin’s two big ones. Martin was still the 6th most effective back on this list when we took more than 40 points off his totals so there is still plenty to like, but with just the one poorly timed dud Foster’s season was probably more useful to his fantasy owners (though given his later ADP Martin’s had more value). Foster was an incredibly steady back, posting less than 8.5 points just once.
  • It’s hard to hold it against him that he was part of Kansas City’s offense and he was employed erratically early in the season without sound explanation, but I’ll admit that this exercise made my reconsider my rank on Jamaal Charles. He posted three duds and is the third lowest per game scorer on this list when controlling for his 30+ point game against the Saints.
  • On the flip side of that conversation, it’s nice to have a reminder of the value of Ray Rice. He posted two duds (5 and 4 point games) against tough opponents but was more or less matchup proof. Maybe he does lose a couple of carries to Bernard Pierce this year, but he’s the third best on the list in controlled average and has been extremely consistent for years. If we’re drafting to avoid floor, Rice is as safe a pick as any and this exercise just adds weight to that notion.
  • CJ Spiller, LeSean McCoy, Steven Jackson and Matt Forte‘s contexts have all changed sufficiently that these numbers are not overly relevant. The latter two had no outliers of concern, anyhow. Nor did McCoy, though his numbers are skewed a bit by missing time to injury last season. Spiller’s first two weeks ended up being his best and his next two were his worst; beyond that he was remarkably consistent with just one week below 10 fantasy points from week five forward.
  • Marshawn Lynch lost one point per game from his average when accounting for a huge week 17, but otherwise was a value to his owners week in and out while posting just two weeks below nine fantasy points. His value is tied to crossing the chalk, but he does so more often than not.
  • Lastly, there are two more second year backs on our list. Alfred Morris is everyone’s darling, but our Mike Omelan warned against him as a bust candidate (relative to ADP at least). His 39.2 fantasy point effort in a playoff clinching game for the Redskins is an anomalous number. Controlling for this output costs Morris 1.5 points on his per game average and yet still his 13.95 ppg average is good for fourth best among the group. Plus he posted just one dud, missing the threshold by 0.1 points. Certainly last year’s production is nothing to cause anyone to shy away. Trent Richardson ranks as more of a risk when considering last year’s numbers, he had three games under six fantasy points and when controlling for his 2TD outlier in just his second career game he moves down a notch from the 6th to 7th highest back on the list. Still, when considering his injury issues and offensive changes he stands to improve on those numbers.

So there you have it – none of this is a reason to take a player off your draft board and if you anchor the position with a back who has a high floor from week-to-week then perhaps a player like Martin with clear week-to-week homerun potential becomes more appealing. Still, you can’t bank on the big weeks. In my running back draft strategy though I’m all about ensuring a safe floor with the early picks and that’s why Arian Foster remains my number two running back as opposed to the trendier picks in Charles and Martin.


Skip to comment form

  1. Trevor

    This is a bit of a flawed piece of methodology here, in my opinion:

    “With that in mind, I took our top 12 consensus ranked running backs and removed any output that was more than ten points greater than their season average (replacing that number in the adjusted formula with their per game average number). I also tallied their ‘duds’ – games in which they scored fewer than six fantasy points.”

    Why replace with their per game average? By doing that, you actually *punish* players for having those big games, rather than just pull the effect of the game out of their data set. The 51.2 point by Martin was an outlier that you can’t rely on, but calling it an average effort in your weighted averages and only assigning him 16.4 points for the week (while a guy like Foster with no outliers gets the full effect of all his best games) is a bit misleading.

    What I would suggest doing to come up with some slightly better numbers is capping those games at your threshold — ie: average game + 10 points. That way they still count as above average games provided by those players, without adding too much extra weight to the overall season average.

    It was easy enough to do myself in the spreadsheet —

    The adjusted averages for Peterson, Martin and Spiller all jump by over a point per week if you give them some credit for having good weeks, rather than replacing them with an average week. Here’s a quick comp:

    Peterson (before): 17.50
    Peterson (after): 18.75

    Martin (before): 13.15
    Martin (after): 14.40

    Foster (unchanged): 16.38

    Charles (before): 11.40
    Charles (after): 12.03

    Lynch (before): 14.45
    Lynch (after): 15.08

    Spiller (before): 11.65
    Spiller (after): 12.9

    McCoy (unchanged): 11.57

    Rice (unchanged): 14.81

    Richardson (before): 12.72
    Richardson (after): 13.39

    Morris (before): 13.95
    Morris (after): 14.58

    Forte (unchanged): 11.09

    Jackson (unchanged): 10.02

    1. Jon Collins

      Hi Trevor,

      Thanks for taking the time to re-run the analysis and provide the feedback. I think you’re right, actually. Now, I used their per-game average before the outliers were taken out of the argument so it isn’t as if this figure is ignored completely; but your point is sound – one of the reasons the excel piece is there and free to use.

      I’d like to point out, though, that adding the 10 points back in for each player changes their adjusted average, but not necessarily their relative value… which is really what my intent was. Spiller, Peterson and Martin all go up by 1.25 under your model, which isn’t really a coincidence it’s just a factor of adding in 10 points divided by number of outliers. Those with one outlier performance all go up by the same figure as their one outlier peers as well.

      So, yes you get a truer number and I appreciate the contribution, genuinely, I wish I had used it… but, I don’t think that the relative values change.

      Please keep coming by. There is always time around here to hear well reasoned thoughts from someone willing to put time and energy into reading and responding!

      1. Trevor

        “I’d like to point out, though, that adding the 10 points back in for each player changes their adjusted average, but not necessarily their relative value…”

        It does change the relative value compared to players with a different number of outliers, though. Martin moves up in comparison to Rice, for example.

        Anyway, I was just using something quick and dirty to change the numbers. To really drill down into outliers, you’d have to start using standard deviations, but that’s a bit more intensive than I wanted to get.

        To be perfectly honest, I’m not all that worried about the relative inconsistency last year in the case of guys like Martin and Spiller.

        Spiller’s variance can probably be written of as 90% a function of usage, which will change completely under the new coaching regime.

        I think the big games by Doug Martin show what he’s capable of, but it shouldn’t be surprising that a rookie running behind a o-line that lost both of its starting guards in the preseason had his ups and downs. It doesn’t seem right to write him off as someone who’s destined to involve a lot more variance than other RBs.

        It’s one thing to chop up the numbers from last year, and say “these numbers meant this player had x value last season.” But things always have to be put in context with the small sample size of football. It seems like it would be daunting to figure out, but I’d be curious to see just how predicative game-to-game variance in one year is when compared to game-to-game variance the next year.

        1. Jon Collins

          Yes. Truthfully I knew I should have calculated standard deviations and I just didn’t have the time when working up the article.

          Great line of inquiry re: is it at all predictive of this year’s performance. Again, not really what I was trying to achieve here… the moral of my story was just that there are additional layers of depth to Martin’s 262.60 fantasy points than just ‘he was 2nd overall last year’ (for example, I’m not meaning to pick on him… for me, very much in conversation as 3rd overall RB).

          I haven’t, and won’t have time to likely, look at YOY outlier predictability – i.e. the likelihood of posting a dud in 2013 because you had three in 2012 (or vice-versa) but I do know that talking about how big games effect per game averages is important. 20+20+20 is an average of 20… so too is 50+5+5… if you think of that as a team’s output as a whole, you probably went 1-2 with the second team.

          Re: that analogy and standard deviation in football in general a great piece appeared on DavidGonos.com earlier; via author Devon Jordan that explains that point more generally (clearly, Trevor, you don’t need the explanation but for those in the audience who are less familiar… http://davidgonos.com/fantasy/football/2013-football/position-depth/)

          1. Jon Collins

            Also, you’re bang on on Spiller. Coaching staff, competition, etc. have changed significantly for him (and for some of the other top 12s) and I said as much in the piece… hard in those instances, for sure, to count on this info being predictive.

Leave a Reply

Your email address will not be published. Required fields are marked *