Advice

RotoViz Staff Projection Accuracy Check: Chase Volume, Not Efficiency

If there is one thing that fantasy football teaches us every year, it’s that the NFL is extremely difficult to predict. We spend months making our RotoViz staff projections only to see LeGarrette Blount score 18 touchdowns and Michael Thomas finish as a top seven wide receiver, not to mention watching a variety of players miss the entire season.1

However, projections are still an important part of preseason analysis. Courtesy of the Projection Machine, the RotoViz staff compiled projections for all 32 teams. With 589 team projections and 347 individual players put under the microscope for 2016, there is no site out there that works harder to give you insight into what we think will happen during the NFL season. Of course, this only matters if we can project with some accuracy. Let’s take a look at how the RotoViz Staff did in 2016.

Overall Accuracy

ScoringPPRPPR/GCut PPRCut PPR/G
Overall0.49940.59070.49730.5729
QB0.49090.45370.25080.1415
RB0.3720.50570.38590.5222
WR0.38070.49340.31290.4043
TE0.42670.52720.30580.452

For every category, I found the correlation coefficient between what our projection was (x) and the result (y). I also broke everything down by position where it was applicable. I wanted to start with our overall projection accuracy for PPR leagues. Here are some of the finer points.

  • QB was our best position in terms of overall scoring, followed by TE, WR, and then RB. I would assume this is due to the amount of missed games at each position.
  • Our overall PPG accuracy absolutely smashed, with an R-squared value of almost 0.6.
  • You can see that our PPG projections for RBs were the second best of the four positions, with QB being the worst.
  • In addition to the overall projections, I made a second group of results based on the players that met certain thresholds.2 Our accuracy was only marginally lower in this group overall, showing just how good our projections were.
  • The one position that completely plummeted with the cutoffs was QB. Once backups were taken out of the equation and we focused on those QBs who had significant playing time, it was much more difficult to be accurate. We saw a similar result with the TEs, though the impact was not as extreme.
  • Our RB projections actually improved with the cutoffs.

Volume

VolumeruATTreTRGpaATTruATT/GreTRG/GpaATT/G
Overall0.6490.48250.49180.73380.59390.4423
QB0.6891N/A0.49180.5135N/A0.4423
RB0.46230.3222N/A0.56310.4954N/A
WR0.11370.4055N/A0.08740.5448N/A
TEN/A0.3649N/AN/A0.4653N/A
  • I looked at our projections for both overall volume as well as the per-game data for each position.
  • We were remarkably accurate at projecting rushing attempts, with an R-squared of 0.649 in total attempts, and a whopping 0.7338 in attempts per game.
  • Our target projections were not as accurate as our rush attempt projections overall. However, our accuracy for RB rush attempts per game and WR targets per game were very close.
  • Something that probably factors into the slightly lower performance in target projection is that we had to first project pass attempts, something we did less accurately than rush attempts overall, and by far the worst on a per game basis.
  • Overall, it would seem that volume is fairly predictable.

Efficiency

EfficiencyruYPCreYPTpaYA
Overall0.00010.09180.006
QB0.4719N/A0.006
RB0.00360.000008N/A
WR0.00060.0658N/A
TEN/A0.0796N/A
  • Projecting efficiency was an absolute nightmare, with none of the three tested statistics even reaching a 0.1 R-squared value.
  • The one area we were relatively succesful at projecting was QB yards per carry. One explanation for this could be that the overall expected range for QB yards per carry is much wider than for the other positions.
  • Josh (@FantasyADHD) recommended in our RotoViz Slack chat that perhaps the best way to project efficiency is with a margin of error. In that way, we could see a range of outcomes instead of a hard guess at efficiency.

Conclusion

Overall it seems as though the RotoViz staff really crushed the projections in 2016. One of my biggest takeaways here was just how much easier it was to project volume than efficiency. It could mean that we just really suck at projecting efficiency, but I think a much more reasonable conclusion is that efficiency is incredibly noisy, while volume we can reasonably project once we have an idea of team success.3 If we can project volume first, and then give a range of outcomes based on efficiency, we should be able to give a nice outlook for players going forward.

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

  1. But Keenan Allen is already sprinting guys! Keenan 2017!  (back)
  2. 25 pass attempts per game at QB, five rush attempts per game at RB, three targets per game at WR, two targets per game at TE, and a six games played minimum across positions.  (back)
  3. Something that can be found rather easily using something like Vegas win totals, though I’m sure there are other methods of projecting this that could potentially be more accurate.  (back)
By Anthony Amico | @amicsta | Archive

Comments   Add comment