Week 5 Air Yards Buy Low WRs

Each week I break out air yards data to attempt to find the wide receivers that stand the best chance to see an increase in their fantasy production.


Air yards give us a complete view of a wide receiver’s game. They show us the volume of yards that are thrown at a guy, the amount of those yards he caught, and how many yards after the catch he created on passes he did catch. Matt Harmon has tweeted out some very cool route visualizations this season from NFL Next Gen Stats. Below is one for Chris Hogan in Week 5.

What I love about these visualizations is that they graphically depict the information captured in air yards.

From the above picture, here are the corresponding air yards data:

Completed Air Yards (white): 101
Incomplete Air Yards (gray): 15
YAC (green): 13
aYPT: 23.2
RACR:1 0.98

Changes in this week’s format

With five weeks of data to work with at this point, the air yards model used to generate buy low recommendations has gotten quite predictive. This is both good and bad. It’s good because moving forward it should give us a solid sense of what players in stable situations will do for the rest of the season. On the other hand the model is blind to the changes taking place in receiver usage in very recent weeks due to injuries and other factors.

To counter this, moving forward I’m going to present two models in this column. One will be a three week model, and the other a 6-8 week version.2 The three week model will move through the season along with us. In other words, the model will be trained on data from the three weeks prior to the week in question.3

Finally, for players that have recently seen large changes in their usage due to injury or otherwise, I’ll profile just the previous week’s air yards numbers at the team level.

Year to date air yards model

Some portion of our readers are interested in predictive modeling, so this is for them. I’m pleased to report the YTD air yards model is fairly powerful, with an out-of-sample r-squared of 0.65. The model is based on two sub-models, both of which are strong in their own right.

The first is a glmnet4 model which combines the benefits of both lasso and ridge regression into one ensemble model. Basically the combination of ridge and lasso penalize certain coefficients and allow an analyst to address many of the requirements implicit in normal multiple linear regression as a matter of course. In addition, I pre-processed the data by removing zero variance predictors,5 centered,6 and scaled7 the data, and then performed principal component analysis.8

The second sub-model is a random forest created using the ranger algorithm. Both models are subject to 10-fold cross validation, and since the models are not highly correlated,9 they were combined into a single “uber” model that is more predictive than either model on its own. The final weighting was roughly 70 percent glmnet and 30 percent ranger. Finally here is a graph of the out of sample fit.


And here are the predictions from the model, along with actual PPR points per game and the difference between the two.

Week 5 YTD Air Yards Model

Terrelle Pryor9.24.850.819.686.667.6137.40.490.
Brandon Marshall10.84.853.818.888.072.6141.80.430.414.814.50.3
Allen Robinson10.35.347.012.590.559.5137.50.420.814.615.7-1.1
A.J. Green10.47.271.831.841.0103.6112.80.410.416.020.0-3.9
Will Fuller8.04.037.827.698.865.4136.60.410.413.212.90.2
Mike Evans12.06.579.510.591.890.0171.30.400.816.820.0-3.2
Jordy Nelson10.05.337.024.087.561.0124.50.391.314.918.9-4.0
Julio Jones8.24.862.840.652.8103.4115.60.390.615.118.7-3.7
Odell Beckham10.25.449.822.059.471.8109.20.390.213.013.8-0.7
Marvin Jones8.25.468.635.234.8103.8103.40.390.614.519.4-4.9
T.Y. Hilton11.07.072.828.650.6101.4123.40.380.615.820.7-4.9
Emmanuel Sanders10.26.458.616.036.874.695.40.380.613.517.5-4.0
Amari Cooper9.
Jeremy Maclin9.35.048.312.853.861.0102.00.360.311.812.6-0.8
Jordan Matthews7.54.845.322.035.867.381.00.350.512.214.5-2.3
Sammie Coates6.23.862.621.652.284.2114.80.350.412.814.6-1.9
Jeremy Kerley9.05.240.820.040.260.881.00.330.411.713.7-2.0
Travis Benjamin7.85.653.025.843.278.896.20.320.412.515.9-3.3
Antonio Brown11.27.466.423.039.689.4106.00.321.015.222.3-7.1
Tajae Sharpe7.23.630.87.061.837.892.60.310.
Chris Hogan4.02.432.614.640.447.273.00.310.
Nelson Agholor5.53.327.39.545.336.872.50.310.
Michael Crabtree8.85.858.612.431.871.090.40.311.012.918.9-6.0
Kenny Stills4.21.829.811.648.
Mike Wallace7.84.240.813.857.854.698.60.310.611.313.3-1.9
Tavon Austin9.04.626.017.643.243.669.20.300.29.710.2-0.5
Marquise Goodwin5.31.525.513.569.035.894.
Demaryius Thomas6.
Alshon Jeffery6.24.459.419.425.878.885.
DeAndre Hopkins8.44.447.49.246.456.693.80.280.611.613.7-2.1
Kevin White9.04.833.313.569.846.8103.
Dez Bryant7.73.743.36.761.050.0104.30.280.311.210.70.5
Kenny Britt6.44.646.824.416.871.
John Brown7.63.838.411.077.449.4115.
Tyrell Williams6.84.240.830.840.471.681.20.270.412.013.8-1.8
DeSean Jackson6.43.637.018.643.655.680.
DeVante Parker6.33.845.516.340.061.885.50.270.310.611.4-0.8
Brandin Cooks8.04.535.528.349.363.884.80.270.511.813.9-2.0
Adam Thielen5.04.042.611.822.454.465.
Torrey Smith4.61.818.62.647.821.
Stefon Diggs8.56.362.031.017.893.
Jarvis Landry9.66.845.
Michael Floyd6.82.429.64.477.434.0107.
Robert Woods6.
Doug Baldwin7.
Davante Adams6.03.334.012.044.846.
Phillip Dorsett4.
Sterling Shepard6.64.438.417.028.455.466.80.240.410.312.3-2.0
Brian Quick3.
Julian Edelman7.64.826.220.027.446.
Kelvin Benjamin7.34.047.512.329.359.576.
Vincent Jackson6.
Brandon LaFell6.
Golden Tate6.23.411.217.846.626.857.
Allen Hurns6.83.831.325.338.856.570.
Chris Conley5.83.832.310.826.343.
Michael Thomas7.85.337.827.326.857.364.50.210.511.514.0-2.4
Cole Beasley6.65.434.232.211.866.446.
Larry Fitzgerald9.66.450.222.835.873.
Eric Decker7.03.054.710.056.364.7111.00.200.711.113.5-2.4
Victor Cruz5.
Breshad Perriman4.
Ted Ginn4.
Corey Coleman6.53.576.010.557.586.5133.
Steve Smith8.05.438.423.622.662.
Quincy Enunwa7.85.437.621.224.258.861.
Terrance Williams4.84.040.518.512.
Charles Johnson3.
Dontrelle Inman4.82.620.613.033.633.654.
Rishard Matthews4.
Pierre Garcon6.44.432.013.620.
Andre Johnson4.21.814.03.034.817.
Jamison Crowder6.24.222.423.823.446.
Randall Cobb7.35.327.533.320.560.
Mohamed Sanu5.23.220.618.423.839.
Willie Snead6.
Marqise Lee6.04.014.827.533.542.348.
Eddie Royal6.
Tyler Boyd4.
Quinton Patton4.02.414.89.617.624.432.
Brice Butler3.21.614.01.415.215.429.
Sammy Watkins5.
Ricardo Louis2.81.610.
Paul Richardson2.31.316.32.020.318.336.
Brandon Coleman5.03.728.
Jaron Brown3.61.421.
Danny Amendola2.
Corey Brown5.03.322.810.019.832.842.
Devin Funchess3.31.312.89.826.
Anquan Boldin5.64.219.617.010.236.829.
Seth Roberts4.42.410.
Cameron Meredith6.75.034.726.019.360.754.00.110.310.013.1-3.1
Malcolm Mitchell3.01.59.511.021.320.530.
Robby Anderson3.31.714.
Chester Rogers3.01.316.80.023.515.840.
Tyler Lockett3.
Andrew Hawkins4.02.413.811.413.625.
Adam Humphries7.34.813.837.526.351.340.
Eli Rogers5.
Dorial Green-Beckham3.52.510.516.59.327.
Aldrick Robinson2.31.320.00.510.020.530.
Jaelen Strong5.02.814.56.818.521.333.
Tyreek Hill3.52.82.814.019.516.822.
Josh Doctson3.
Greg Salas3.
Taylor Gabriel3.02.729.
Jermaine Kearse4.
Markus Wheaton3.01.314.32.723.717.
Bennie Fowler2.
Kamar Aiken2.
Albert Wilson5.33.39.711.714.021.323.
Donte Moncrief5.03.528.
Chris Moore2.
Walt Powell3.
Jordan Norwood2.
Darrius Heyward-Bey1.
J.J. Nelson1.30.711.30.321.711.733.
Rashad Greene1.
Matthew Slater0.
Justin Hardy1.
Jalin Marshall2.71.714.
Rod Streater1.
Harry Douglas4.
Cody Latimer1.
Andre Holmes1.
Keenan Allen7.
Kendall Wright2.
Tommylee Lewis2.01.710.07.02.317.
Tanner McEvoy1.00.516.05.00.521.
Aaron Burbridge0.
Cordarrelle Patterson2.82.22.615.63.618.
Rashard Higgins1.
Trevor Davis0.
Andre Roberts0.
Ryan Grant1.
Tavarres King1.
Jarius Wright4.
Quan Bray0.
Leonte Carroo1.
Cecil Shorts2.
Charone Peake2.
Russell Shepard1.
Alex Erickson0.
Jordan Taylor1.
Braxton Miller2.
Arrelious Benn1.
Bryan Walters1.
Josh Bellamy0.
Pharoh Cooper1.
James Wright1.
Bradley Marquez1.
Dexter McCluster2.01.3-2.710.
Jared Abbrederis1.
Rashad Ross1.
Mike Thomas0.
Justin Hunter1.

Three week rolling air yards model

I ran the same model on data from just weeks 3-5 from 2009-2015. The idea here is to try and capture changes in receiver usage that might be papered over by using full season averages. Below is the list of top buy-low candidates for this period.10

full_name comp_air_yds incomp_air_yds yac air_yds ms_air_yds predict actual dif
Robert Woods 42.0 46.7 13.3 88.7 0.34 11.4 10.5 0.9
Mike Wallace 39.0 60.3 8.0 99.3 0.32 11.1 9.4 1.8
Will Fuller 18.7 73.7 20.0 92.3 0.30 10.3 9.5 0.8
Michael Floyd 31.7 86.7 2.0 118.3 0.26 10.2 7.7 2.5
Quincy Enunwa 27.3 39.7 22.0 67.0 0.19 10.0 9.6 0.4
DeAndre Hopkins 32.0 50.3 6.7 82.3 0.27 9.9 9.2 0.7
Brandon Coleman 36.0 18.0 11.5 54.0 0.18 9.6 9.3 0.4
Ted Ginn 32.5 74.5 5.5 107.0 0.28 9.4 6.3 3.1
Vincent Jackson 37.0 46.0 0.5 83.0 0.20 8.8 6.8 2.1
Marqise Lee 7.5 44.0 28.5 51.5 0.19 8.7 8.1 0.6
Chris Hogan 32.7 48.7 6.3 81.3 0.36 8.5 5.6 3.0
Tajae Sharpe 21.0 65.3 5.7 86.3 0.28 8.1 5.0 3.1

Week 5 Only Team Data



With Stefon Diggs out with a groin injury, Adam Thielen saw 63 percent of the Vikings air yards and a 23 percent target share. He is a buy. Cordarelle Patterson, with just 12 percent of the team’s air yards and less YAC than Theielen, is not.



In Week 5 Sammy Coates seemed to seize the mantle of stretch X that has been unclaimed in the Steelers offense since Martavis Bryant was suspended. Not only did Coates lead the team in market share of air yards, he also tied Antonio Brown in target share. Coates’ success should be good news for all the other offensive players on Pittsburgh, but it is too soon to call him a solid buy.



Week 5 might be a bad sign for Jordan Mathews moving forward. Nelson Agholor not only led the team in MS Air but also in target share. On the other hand, Agholor’s RACR is almost irredeemably low. He added no YAC and caught only 30 percent of the passes thrown his way. Monitor Matthews usage this week and hope for a bounce back in volume.

Full Week 5 Air Yards Data Dump

Week 5 Only Air Yards

Mike EvansTB61211881392278903480.6518.90.390.500.40
Adam ThielenMIN78131963312912702050.6316.10.980.880.27
Brandon MarshallNYJ815119956716211402600.6210.80.700.530.39
Jordy NelsonGB413114242022263804060.5617.40.170.310.29
Sammie CoatesPIT611253867716313903230.5014.80.850.550.23
Nelson AgholorPHI270225951202752410.5017.
Amari CooperOAK612150887516313803330.4913.60.850.500.30
DeVante ParkerMIA23018524567001240.4518.71.250.670.17
Travis BenjaminSD7110457213921111704700.4519.20.550.640.37
Jeremy KerleySF813125773711410202660.438.80.890.620.42
T.Y. HiltonIND10111361351414917103480.4313.51.150.910.28
Chris HoganNE450131011511611422790.4223.20.980.800.12
A.J. GreenCIN4801139961355003440.3916.90.370.500.20
Robert WoodsBUF26052150712601810.3911.80.370.330.26
Terrance WilliamsDAL55017530537001400.3810.61.321.000.21
Brian QuickLA340163538735101940.3818.30.700.750.13
Julio JonesATL260191081912902560.3615.20.320.330.21
Odell BeckhamNYG5121223452865602560.347.20.650.420.34
Demaryius ThomasDEN5712425791044903110.3314.90.470.710.20
Ricardo LouisCLE2509483871302640.3317.40.150.400.16
Golden TateDET35028111829396900.325.81.340.600.20
Marvin JonesDET4511324529370900.325.81.280.800.20
Tajae SharpeTEN2408975841702630.3221.00.200.500.14
Emmanuel SandersDEN79097126978003110.3110.80.820.780.26
Anquan BoldinDET4402028028480900.317.01.711.000.16
Terrelle PryorCLE560282458824832640.3113.70.590.830.19
Kelvin BenjaminCAR59046620867002780.319.60.810.560.32
Sterling ShepardNYG2707772791402560.3111.
Will FullerHOU1609-510499403290.3016.
Cameron MeredithCHI91215971249513063290.297.91.370.750.28
Breshad PerrimanBAL2502983921103230.2818.40.120.400.11
Tyrell WilliamsSD561101072413111704700.2821.80.890.830.20
Brandon LaFellCIN8112194946956803440.288.60.720.730.27
Pierre GarconWAS57125430845603070.2712.00.670.710.17
Mike WallaceBAL7110115235876303230.277.90.720.640.24
Kenny BrittLA560413416507501940.268.31.500.830.19
Michael CrabtreeOAK37104735824703330.2511.70.570.430.18
Tavon AustinLA7100174254759261940.
Julian EdelmanNE5100152047673502790.246.70.520.500.24
Chester RogersIND36002458822403480.2413.70.290.500.15
Kenny StillsMIA010002929001240.2329.
Davante AdamsGB581275835938504060.2311.60.910.630.18
DeAndre HopkinsHOU591183836745603290.228.20.760.560.21
Alshon JefferyCHI560225519747703290.2212.31.040.830.14
Andrew HawkinsCLE471144217595602640.228.40.950.570.22
Antonio BrownPIT9111225616727803230.
DeSean JacksonWAS370122345683503070.229.70.510.430.17
Larry FitzgeraldARI68216659748103400.
John BrownARI1408371741103400.2218.
Jordan MatthewsPHI44013520526502410.2213.
Cole BeasleyDAL44123300305301400.217.51.771.000.17
Jaelen StrongHOU590152444683903290.217.60.570.560.21
Andre JohnsonTEN24101340531302630.2013.30.250.500.14
Quincy EnunwaNYJ470262524495102600.
Mohamed SanuATL36073612484302560.198.00.900.500.21
Michael FloydARI030005959003400.1719.
Marquise GoodwinBUF26121120311301810.175.20.420.330.26
Eddie RoyalCHI790152828564303290.176.20.770.780.21
Tyler BoydCIN35003324573303440.1711.40.580.600.12
Robby AndersonNYJ12001033431002600.1721.50.230.500.05
Torrey SmithSF010004343002660.1643.
Jarvis LandryMIA3308200202821240.166.71.401.000.17
Corey BrownCAR24001331441302780.1611.00.300.500.14
Phillip DorsettIND23012528532603480.1517.70.490.670.08
Vincent JacksonTB35063023533603480.1510.60.680.600.17
Dontrelle InmanSD130126769304700.1523.00.040.330.10
Markus WheatonPIT23081136471903230.1515.70.400.670.06
Brice ButlerDAL2300200202001400.
Jaron BrownARI030004747003400.1415.
J.J. NelsonARI010004646003400.1446.
Randall CobbGB91107236155110804060.
Cordarrelle PattersonMIN461271212243972050.124.01.630.670.20
Victor CruzNYG020002828002560.1114.
Rod StreaterSF010002828002660.1128.
Rishard MatthewsTEN4415270273202630.
Tavarres KingNYG010002626002560.1026.
Jarius WrightMIN44012200203202050.105.01.601.000.13
Kamar AikenBAL030003131003230.1010.
Steve SmithBAL3302270272903230.
Jamison CrowderWAS33010250253503070.088.31.401.000.07
Dorial Green-BeckhamPHI34039414184302410.074.52.390.750.12
Andre HolmesOAK23001310231303330.077.70.570.670.08
Andre RobertsDET010006600900.
Seth RobertsOAK2302201212203330.
Quan BrayIND23013163192973480.056.31.530.670.08
Adam HumphriesTB1209217191103480.059.50.580.500.07
Pharoh CooperLA010001010001940.0510.
Aldrick RobinsonATL010001313002560.0513.
Jordan NorwoodDEN23013113142403110.054.71.710.670.09
Bennie FowlerDEN2200130131303110.
Quinton PattonSF2205110111602660.045.51.451.000.06
Josh BellamyCHI1101130131403290.0413.
Malcolm MitchellNE23016410702790.043.30.700.670.07
Walt PowellBUF0100066001810.
Rashad RossWAS1101707803070.
Justin HunterBUF1110404401810.
Kendall WrightTEN1104505902630.025.01.801.000.03
Charone PeakeNYJ1105101602600.
  1. RACR is a measure of how well a receiver converts air yards into receiving yards. The formula is: (completed air yards + YAC) / total air yards. It combines a receiver’s catching ability with his ability to create yards after the catch into one efficiency metric. It is the most predictive efficiency metric I have found for WRs.  (back)
  2. I haven’t decided yet  (back)
  3. So for Week 9 for example, the three week model will have been trained on data from weeks 6-8 from 2009-2015  (back)
  4. Also called elasticnet.  (back)
  5. These add no information to the model so are best removed.  (back)
  6. The mean is subtracted from each of the values.  (back)
  7. Each value is divided by the standard deviation.  (back)
  8. These pre-processing steps are a very common recipe in machine learning modeling.  (back)
  9. -0.31 correlation  (back)
  10. The out of sample r-squared for the model is 0.50.  (back)

Josh Hermsmeyer

Air yards, Numbers Game, and predictive modeling.
What we do


Sign-up today for our free Premium Email subscription!

© 2019 RotoViz. All rights Reserved.